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" 1612aa4082SMax Reitz #include "qemu/coroutine.h" 171181e19aSMax Reitz #include "qemu/range.h" 18893f7ebaSPaolo Bonzini #include "trace.h" 19c87621eaSJohn Snow #include "block/blockjob_int.h" 20737e150eSPaolo Bonzini #include "block/block_int.h" 21373340b2SMax Reitz #include "sysemu/block-backend.h" 22da34e65cSMarkus Armbruster #include "qapi/error.h" 23cc7a8ea7SMarkus Armbruster #include "qapi/qmp/qerror.h" 24893f7ebaSPaolo Bonzini #include "qemu/ratelimit.h" 25b812f671SPaolo Bonzini #include "qemu/bitmap.h" 26893f7ebaSPaolo Bonzini 27402a4741SPaolo Bonzini #define MAX_IN_FLIGHT 16 28b436982fSEric Blake #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ 29b436982fSEric Blake #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) 30402a4741SPaolo Bonzini 31402a4741SPaolo Bonzini /* The mirroring buffer is a list of granularity-sized chunks. 32402a4741SPaolo Bonzini * Free chunks are organized in a list. 33402a4741SPaolo Bonzini */ 34402a4741SPaolo Bonzini typedef struct MirrorBuffer { 35402a4741SPaolo Bonzini QSIMPLEQ_ENTRY(MirrorBuffer) next; 36402a4741SPaolo Bonzini } MirrorBuffer; 37893f7ebaSPaolo Bonzini 3812aa4082SMax Reitz typedef struct MirrorOp MirrorOp; 3912aa4082SMax Reitz 40893f7ebaSPaolo Bonzini typedef struct MirrorBlockJob { 41893f7ebaSPaolo Bonzini BlockJob common; 42e253f4b8SKevin Wolf BlockBackend *target; 434ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 445bc361b8SFam Zheng BlockDriverState *base; 454ef85a9cSKevin Wolf 4609158f00SBenoît Canet /* The name of the graph node to replace */ 4709158f00SBenoît Canet char *replaces; 4809158f00SBenoît Canet /* The BDS to replace */ 4909158f00SBenoît Canet BlockDriverState *to_replace; 5009158f00SBenoît Canet /* Used to block operations on the drive-mirror-replace target */ 5109158f00SBenoît Canet Error *replace_blocker; 5203544a6eSFam Zheng bool is_none_mode; 53274fcceeSMax Reitz BlockMirrorBackingMode backing_mode; 54d06107adSMax Reitz MirrorCopyMode copy_mode; 55b952b558SPaolo Bonzini BlockdevOnError on_source_error, on_target_error; 56d63ffd87SPaolo Bonzini bool synced; 57d06107adSMax Reitz /* Set when the target is synced (dirty bitmap is clean, nothing 58d06107adSMax Reitz * in flight) and the job is running in active mode */ 59d06107adSMax Reitz bool actively_synced; 60d63ffd87SPaolo Bonzini bool should_complete; 61eee13dfeSPaolo Bonzini int64_t granularity; 62b812f671SPaolo Bonzini size_t buf_size; 63b21c7652SMax Reitz int64_t bdev_length; 64b812f671SPaolo Bonzini unsigned long *cow_bitmap; 65e4654d2dSFam Zheng BdrvDirtyBitmap *dirty_bitmap; 66dc162c8eSFam Zheng BdrvDirtyBitmapIter *dbi; 67893f7ebaSPaolo Bonzini uint8_t *buf; 68402a4741SPaolo Bonzini QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; 69402a4741SPaolo Bonzini int buf_free_count; 70bd48bde8SPaolo Bonzini 7149efb1f5SDenis V. Lunev uint64_t last_pause_ns; 72402a4741SPaolo Bonzini unsigned long *in_flight_bitmap; 73bd48bde8SPaolo Bonzini int in_flight; 74b436982fSEric Blake int64_t bytes_in_flight; 75b58deb34SPaolo Bonzini QTAILQ_HEAD(, MirrorOp) ops_in_flight; 76bd48bde8SPaolo Bonzini int ret; 770fc9f8eaSFam Zheng bool unmap; 78b436982fSEric Blake int target_cluster_size; 79e5b43573SFam Zheng int max_iov; 8090ab48ebSAnton Nefedov bool initial_zeroing_ongoing; 81d06107adSMax Reitz int in_active_write_counter; 82737efc1eSJohn Snow bool prepared; 83893f7ebaSPaolo Bonzini } MirrorBlockJob; 84893f7ebaSPaolo Bonzini 85429076e8SMax Reitz typedef struct MirrorBDSOpaque { 86429076e8SMax Reitz MirrorBlockJob *job; 87429076e8SMax Reitz } MirrorBDSOpaque; 88429076e8SMax Reitz 8912aa4082SMax Reitz struct MirrorOp { 90bd48bde8SPaolo Bonzini MirrorBlockJob *s; 91bd48bde8SPaolo Bonzini QEMUIOVector qiov; 92b436982fSEric Blake int64_t offset; 93b436982fSEric Blake uint64_t bytes; 942e1990b2SMax Reitz 952e1990b2SMax Reitz /* The pointee is set by mirror_co_read(), mirror_co_zero(), and 962e1990b2SMax Reitz * mirror_co_discard() before yielding for the first time */ 972e1990b2SMax Reitz int64_t *bytes_handled; 9812aa4082SMax Reitz 991181e19aSMax Reitz bool is_pseudo_op; 100d06107adSMax Reitz bool is_active_write; 10112aa4082SMax Reitz CoQueue waiting_requests; 10212aa4082SMax Reitz 10312aa4082SMax Reitz QTAILQ_ENTRY(MirrorOp) next; 10412aa4082SMax Reitz }; 105bd48bde8SPaolo Bonzini 1064295c5fcSMax Reitz typedef enum MirrorMethod { 1074295c5fcSMax Reitz MIRROR_METHOD_COPY, 1084295c5fcSMax Reitz MIRROR_METHOD_ZERO, 1094295c5fcSMax Reitz MIRROR_METHOD_DISCARD, 1104295c5fcSMax Reitz } MirrorMethod; 1114295c5fcSMax Reitz 112b952b558SPaolo Bonzini static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 113b952b558SPaolo Bonzini int error) 114b952b558SPaolo Bonzini { 115b952b558SPaolo Bonzini s->synced = false; 116d06107adSMax Reitz s->actively_synced = false; 117b952b558SPaolo Bonzini if (read) { 11881e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_source_error, 11981e254dcSKevin Wolf true, error); 120b952b558SPaolo Bonzini } else { 12181e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_target_error, 12281e254dcSKevin Wolf false, error); 123b952b558SPaolo Bonzini } 124b952b558SPaolo Bonzini } 125b952b558SPaolo Bonzini 1261181e19aSMax Reitz static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, 1271181e19aSMax Reitz MirrorBlockJob *s, 1281181e19aSMax Reitz uint64_t offset, 1291181e19aSMax Reitz uint64_t bytes) 1301181e19aSMax Reitz { 1311181e19aSMax Reitz uint64_t self_start_chunk = offset / s->granularity; 1321181e19aSMax Reitz uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); 1331181e19aSMax Reitz uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; 1341181e19aSMax Reitz 1351181e19aSMax Reitz while (find_next_bit(s->in_flight_bitmap, self_end_chunk, 1361181e19aSMax Reitz self_start_chunk) < self_end_chunk && 1371181e19aSMax Reitz s->ret >= 0) 1381181e19aSMax Reitz { 1391181e19aSMax Reitz MirrorOp *op; 1401181e19aSMax Reitz 1411181e19aSMax Reitz QTAILQ_FOREACH(op, &s->ops_in_flight, next) { 1421181e19aSMax Reitz uint64_t op_start_chunk = op->offset / s->granularity; 1431181e19aSMax Reitz uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, 1441181e19aSMax Reitz s->granularity) - 1451181e19aSMax Reitz op_start_chunk; 1461181e19aSMax Reitz 1471181e19aSMax Reitz if (op == self) { 1481181e19aSMax Reitz continue; 1491181e19aSMax Reitz } 1501181e19aSMax Reitz 1511181e19aSMax Reitz if (ranges_overlap(self_start_chunk, self_nb_chunks, 1521181e19aSMax Reitz op_start_chunk, op_nb_chunks)) 1531181e19aSMax Reitz { 1541181e19aSMax Reitz qemu_co_queue_wait(&op->waiting_requests, NULL); 1551181e19aSMax Reitz break; 1561181e19aSMax Reitz } 1571181e19aSMax Reitz } 1581181e19aSMax Reitz } 1591181e19aSMax Reitz } 1601181e19aSMax Reitz 1612e1990b2SMax Reitz static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) 162bd48bde8SPaolo Bonzini { 163bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 164402a4741SPaolo Bonzini struct iovec *iov; 165bd48bde8SPaolo Bonzini int64_t chunk_num; 166b436982fSEric Blake int i, nb_chunks; 167bd48bde8SPaolo Bonzini 168b436982fSEric Blake trace_mirror_iteration_done(s, op->offset, op->bytes, ret); 169bd48bde8SPaolo Bonzini 170bd48bde8SPaolo Bonzini s->in_flight--; 171b436982fSEric Blake s->bytes_in_flight -= op->bytes; 172402a4741SPaolo Bonzini iov = op->qiov.iov; 173402a4741SPaolo Bonzini for (i = 0; i < op->qiov.niov; i++) { 174402a4741SPaolo Bonzini MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 175402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 176402a4741SPaolo Bonzini s->buf_free_count++; 177402a4741SPaolo Bonzini } 178402a4741SPaolo Bonzini 179b436982fSEric Blake chunk_num = op->offset / s->granularity; 180b436982fSEric Blake nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 18112aa4082SMax Reitz 182402a4741SPaolo Bonzini bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 18312aa4082SMax Reitz QTAILQ_REMOVE(&s->ops_in_flight, op, next); 184b21c7652SMax Reitz if (ret >= 0) { 185b21c7652SMax Reitz if (s->cow_bitmap) { 186bd48bde8SPaolo Bonzini bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 187bd48bde8SPaolo Bonzini } 18890ab48ebSAnton Nefedov if (!s->initial_zeroing_ongoing) { 18930a5c887SKevin Wolf job_progress_update(&s->common.job, op->bytes); 190b21c7652SMax Reitz } 19190ab48ebSAnton Nefedov } 1926df3bf8eSZhang Min qemu_iovec_destroy(&op->qiov); 1937b770c72SStefan Hajnoczi 19412aa4082SMax Reitz qemu_co_queue_restart_all(&op->waiting_requests); 19512aa4082SMax Reitz g_free(op); 1967b770c72SStefan Hajnoczi } 197bd48bde8SPaolo Bonzini 1982e1990b2SMax Reitz static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) 199bd48bde8SPaolo Bonzini { 200bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 201b9e413ddSPaolo Bonzini 202bd48bde8SPaolo Bonzini if (ret < 0) { 203bd48bde8SPaolo Bonzini BlockErrorAction action; 204bd48bde8SPaolo Bonzini 205e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 206bd48bde8SPaolo Bonzini action = mirror_error_action(s, false, -ret); 207a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 208bd48bde8SPaolo Bonzini s->ret = ret; 209bd48bde8SPaolo Bonzini } 210bd48bde8SPaolo Bonzini } 211d12ade57SVladimir Sementsov-Ogievskiy 212bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 213bd48bde8SPaolo Bonzini } 214bd48bde8SPaolo Bonzini 2152e1990b2SMax Reitz static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) 216bd48bde8SPaolo Bonzini { 217bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 218b9e413ddSPaolo Bonzini 219bd48bde8SPaolo Bonzini if (ret < 0) { 220bd48bde8SPaolo Bonzini BlockErrorAction action; 221bd48bde8SPaolo Bonzini 222e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 223bd48bde8SPaolo Bonzini action = mirror_error_action(s, true, -ret); 224a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 225bd48bde8SPaolo Bonzini s->ret = ret; 226bd48bde8SPaolo Bonzini } 227bd48bde8SPaolo Bonzini 228bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 229d12ade57SVladimir Sementsov-Ogievskiy return; 230bd48bde8SPaolo Bonzini } 231d12ade57SVladimir Sementsov-Ogievskiy 232d12ade57SVladimir Sementsov-Ogievskiy ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0); 233d12ade57SVladimir Sementsov-Ogievskiy mirror_write_complete(op, ret); 234b9e413ddSPaolo Bonzini } 235bd48bde8SPaolo Bonzini 236782d97efSEric Blake /* Clip bytes relative to offset to not exceed end-of-file */ 237782d97efSEric Blake static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, 238782d97efSEric Blake int64_t offset, 239782d97efSEric Blake int64_t bytes) 240782d97efSEric Blake { 241782d97efSEric Blake return MIN(bytes, s->bdev_length - offset); 242782d97efSEric Blake } 243782d97efSEric Blake 244782d97efSEric Blake /* Round offset and/or bytes to target cluster if COW is needed, and 245782d97efSEric Blake * return the offset of the adjusted tail against original. */ 246782d97efSEric Blake static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, 247ae4cc877SEric Blake uint64_t *bytes) 248893f7ebaSPaolo Bonzini { 249e5b43573SFam Zheng bool need_cow; 250e5b43573SFam Zheng int ret = 0; 251782d97efSEric Blake int64_t align_offset = *offset; 2527cfd5275SEric Blake int64_t align_bytes = *bytes; 253782d97efSEric Blake int max_bytes = s->granularity * s->max_iov; 254893f7ebaSPaolo Bonzini 255782d97efSEric Blake need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); 256782d97efSEric Blake need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, 257e5b43573SFam Zheng s->cow_bitmap); 258e5b43573SFam Zheng if (need_cow) { 259782d97efSEric Blake bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, 260782d97efSEric Blake &align_offset, &align_bytes); 2618f0720ecSPaolo Bonzini } 2628f0720ecSPaolo Bonzini 263782d97efSEric Blake if (align_bytes > max_bytes) { 264782d97efSEric Blake align_bytes = max_bytes; 265e5b43573SFam Zheng if (need_cow) { 266782d97efSEric Blake align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); 267e5b43573SFam Zheng } 268e5b43573SFam Zheng } 269782d97efSEric Blake /* Clipping may result in align_bytes unaligned to chunk boundary, but 2704150ae60SFam Zheng * that doesn't matter because it's already the end of source image. */ 271782d97efSEric Blake align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); 272402a4741SPaolo Bonzini 273782d97efSEric Blake ret = align_offset + align_bytes - (*offset + *bytes); 274782d97efSEric Blake *offset = align_offset; 275782d97efSEric Blake *bytes = align_bytes; 276e5b43573SFam Zheng assert(ret >= 0); 277e5b43573SFam Zheng return ret; 278e5b43573SFam Zheng } 279e5b43573SFam Zheng 280537c3d4fSStefan Hajnoczi static inline void coroutine_fn 281537c3d4fSStefan Hajnoczi mirror_wait_for_any_operation(MirrorBlockJob *s, bool active) 28221cd917fSFam Zheng { 28312aa4082SMax Reitz MirrorOp *op; 28412aa4082SMax Reitz 2851181e19aSMax Reitz QTAILQ_FOREACH(op, &s->ops_in_flight, next) { 2861181e19aSMax Reitz /* Do not wait on pseudo ops, because it may in turn wait on 2871181e19aSMax Reitz * some other operation to start, which may in fact be the 2881181e19aSMax Reitz * caller of this function. Since there is only one pseudo op 2891181e19aSMax Reitz * at any given time, we will always find some real operation 2901181e19aSMax Reitz * to wait on. */ 291d06107adSMax Reitz if (!op->is_pseudo_op && op->is_active_write == active) { 29212aa4082SMax Reitz qemu_co_queue_wait(&op->waiting_requests, NULL); 2931181e19aSMax Reitz return; 2941181e19aSMax Reitz } 2951181e19aSMax Reitz } 2961181e19aSMax Reitz abort(); 29721cd917fSFam Zheng } 29821cd917fSFam Zheng 299537c3d4fSStefan Hajnoczi static inline void coroutine_fn 300537c3d4fSStefan Hajnoczi mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) 301d06107adSMax Reitz { 302d06107adSMax Reitz /* Only non-active operations use up in-flight slots */ 303d06107adSMax Reitz mirror_wait_for_any_operation(s, false); 304d06107adSMax Reitz } 305d06107adSMax Reitz 3062e1990b2SMax Reitz /* Perform a mirror copy operation. 3072e1990b2SMax Reitz * 3082e1990b2SMax Reitz * *op->bytes_handled is set to the number of bytes copied after and 3092e1990b2SMax Reitz * including offset, excluding any bytes copied prior to offset due 3102e1990b2SMax Reitz * to alignment. This will be op->bytes if no alignment is necessary, 3112e1990b2SMax Reitz * or (new_end - op->offset) if the tail is rounded up or down due to 312e5b43573SFam Zheng * alignment or buffer limit. 313402a4741SPaolo Bonzini */ 3142e1990b2SMax Reitz static void coroutine_fn mirror_co_read(void *opaque) 315e5b43573SFam Zheng { 3162e1990b2SMax Reitz MirrorOp *op = opaque; 3172e1990b2SMax Reitz MirrorBlockJob *s = op->s; 318ae4cc877SEric Blake int nb_chunks; 319ae4cc877SEric Blake uint64_t ret; 320ae4cc877SEric Blake uint64_t max_bytes; 321402a4741SPaolo Bonzini 322ae4cc877SEric Blake max_bytes = s->granularity * s->max_iov; 323e5b43573SFam Zheng 324e5b43573SFam Zheng /* We can only handle as much as buf_size at a time. */ 3252e1990b2SMax Reitz op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); 3262e1990b2SMax Reitz assert(op->bytes); 3272e1990b2SMax Reitz assert(op->bytes < BDRV_REQUEST_MAX_BYTES); 3282e1990b2SMax Reitz *op->bytes_handled = op->bytes; 329e5b43573SFam Zheng 330e5b43573SFam Zheng if (s->cow_bitmap) { 3312e1990b2SMax Reitz *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); 332e5b43573SFam Zheng } 3332e1990b2SMax Reitz /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ 3342e1990b2SMax Reitz assert(*op->bytes_handled <= UINT_MAX); 3352e1990b2SMax Reitz assert(op->bytes <= s->buf_size); 336ae4cc877SEric Blake /* The offset is granularity-aligned because: 337e5b43573SFam Zheng * 1) Caller passes in aligned values; 338e5b43573SFam Zheng * 2) mirror_cow_align is used only when target cluster is larger. */ 3392e1990b2SMax Reitz assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); 340ae4cc877SEric Blake /* The range is sector-aligned, since bdrv_getlength() rounds up. */ 3412e1990b2SMax Reitz assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); 3422e1990b2SMax Reitz nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 343e5b43573SFam Zheng 344e5b43573SFam Zheng while (s->buf_free_count < nb_chunks) { 3452e1990b2SMax Reitz trace_mirror_yield_in_flight(s, op->offset, s->in_flight); 3461181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 347b812f671SPaolo Bonzini } 348b812f671SPaolo Bonzini 349402a4741SPaolo Bonzini /* Now make a QEMUIOVector taking enough granularity-sized chunks 350402a4741SPaolo Bonzini * from s->buf_free. 351402a4741SPaolo Bonzini */ 352402a4741SPaolo Bonzini qemu_iovec_init(&op->qiov, nb_chunks); 353402a4741SPaolo Bonzini while (nb_chunks-- > 0) { 354402a4741SPaolo Bonzini MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 3552e1990b2SMax Reitz size_t remaining = op->bytes - op->qiov.size; 3565a0f6fd5SKevin Wolf 357402a4741SPaolo Bonzini QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 358402a4741SPaolo Bonzini s->buf_free_count--; 3595a0f6fd5SKevin Wolf qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 360402a4741SPaolo Bonzini } 361402a4741SPaolo Bonzini 362893f7ebaSPaolo Bonzini /* Copy the dirty cluster. */ 363bd48bde8SPaolo Bonzini s->in_flight++; 3642e1990b2SMax Reitz s->bytes_in_flight += op->bytes; 3652e1990b2SMax Reitz trace_mirror_one_iteration(s, op->offset, op->bytes); 366dcfb3bebSFam Zheng 367138f9fffSMax Reitz ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, 368138f9fffSMax Reitz &op->qiov, 0); 3692e1990b2SMax Reitz mirror_read_complete(op, ret); 370e5b43573SFam Zheng } 371e5b43573SFam Zheng 3722e1990b2SMax Reitz static void coroutine_fn mirror_co_zero(void *opaque) 373e5b43573SFam Zheng { 3742e1990b2SMax Reitz MirrorOp *op = opaque; 3752e1990b2SMax Reitz int ret; 376e5b43573SFam Zheng 3772e1990b2SMax Reitz op->s->in_flight++; 3782e1990b2SMax Reitz op->s->bytes_in_flight += op->bytes; 3792e1990b2SMax Reitz *op->bytes_handled = op->bytes; 380e5b43573SFam Zheng 3812e1990b2SMax Reitz ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, 3822e1990b2SMax Reitz op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); 3832e1990b2SMax Reitz mirror_write_complete(op, ret); 384e5b43573SFam Zheng } 3852e1990b2SMax Reitz 3862e1990b2SMax Reitz static void coroutine_fn mirror_co_discard(void *opaque) 3872e1990b2SMax Reitz { 3882e1990b2SMax Reitz MirrorOp *op = opaque; 3892e1990b2SMax Reitz int ret; 3902e1990b2SMax Reitz 3912e1990b2SMax Reitz op->s->in_flight++; 3922e1990b2SMax Reitz op->s->bytes_in_flight += op->bytes; 3932e1990b2SMax Reitz *op->bytes_handled = op->bytes; 3942e1990b2SMax Reitz 3952e1990b2SMax Reitz ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); 3962e1990b2SMax Reitz mirror_write_complete(op, ret); 397e5b43573SFam Zheng } 398e5b43573SFam Zheng 3994295c5fcSMax Reitz static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, 4004295c5fcSMax Reitz unsigned bytes, MirrorMethod mirror_method) 4014295c5fcSMax Reitz { 4022e1990b2SMax Reitz MirrorOp *op; 4032e1990b2SMax Reitz Coroutine *co; 4042e1990b2SMax Reitz int64_t bytes_handled = -1; 4052e1990b2SMax Reitz 4062e1990b2SMax Reitz op = g_new(MirrorOp, 1); 4072e1990b2SMax Reitz *op = (MirrorOp){ 4082e1990b2SMax Reitz .s = s, 4092e1990b2SMax Reitz .offset = offset, 4102e1990b2SMax Reitz .bytes = bytes, 4112e1990b2SMax Reitz .bytes_handled = &bytes_handled, 4122e1990b2SMax Reitz }; 41312aa4082SMax Reitz qemu_co_queue_init(&op->waiting_requests); 4142e1990b2SMax Reitz 4154295c5fcSMax Reitz switch (mirror_method) { 4164295c5fcSMax Reitz case MIRROR_METHOD_COPY: 4172e1990b2SMax Reitz co = qemu_coroutine_create(mirror_co_read, op); 4182e1990b2SMax Reitz break; 4194295c5fcSMax Reitz case MIRROR_METHOD_ZERO: 4202e1990b2SMax Reitz co = qemu_coroutine_create(mirror_co_zero, op); 4212e1990b2SMax Reitz break; 4224295c5fcSMax Reitz case MIRROR_METHOD_DISCARD: 4232e1990b2SMax Reitz co = qemu_coroutine_create(mirror_co_discard, op); 4242e1990b2SMax Reitz break; 4254295c5fcSMax Reitz default: 4264295c5fcSMax Reitz abort(); 4274295c5fcSMax Reitz } 4282e1990b2SMax Reitz 42912aa4082SMax Reitz QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); 4302e1990b2SMax Reitz qemu_coroutine_enter(co); 4312e1990b2SMax Reitz /* At this point, ownership of op has been moved to the coroutine 4322e1990b2SMax Reitz * and the object may already be freed */ 4332e1990b2SMax Reitz 4342e1990b2SMax Reitz /* Assert that this value has been set */ 4352e1990b2SMax Reitz assert(bytes_handled >= 0); 4362e1990b2SMax Reitz 4372e1990b2SMax Reitz /* Same assertion as in mirror_co_read() (and for mirror_co_read() 4382e1990b2SMax Reitz * and mirror_co_discard(), bytes_handled == op->bytes, which 4392e1990b2SMax Reitz * is the @bytes parameter given to this function) */ 4402e1990b2SMax Reitz assert(bytes_handled <= UINT_MAX); 4412e1990b2SMax Reitz return bytes_handled; 4424295c5fcSMax Reitz } 4434295c5fcSMax Reitz 444e5b43573SFam Zheng static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) 445e5b43573SFam Zheng { 446138f9fffSMax Reitz BlockDriverState *source = s->mirror_top_bs->backing->bs; 4471181e19aSMax Reitz MirrorOp *pseudo_op; 4481181e19aSMax Reitz int64_t offset; 4491181e19aSMax Reitz uint64_t delay_ns = 0, ret = 0; 450e5b43573SFam Zheng /* At least the first dirty chunk is mirrored in one iteration. */ 451e5b43573SFam Zheng int nb_chunks = 1; 4524b5004d9SDenis V. Lunev bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); 453b436982fSEric Blake int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); 454e5b43573SFam Zheng 455b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_lock(s->dirty_bitmap); 456f798184cSEric Blake offset = bdrv_dirty_iter_next(s->dbi); 457fb2ef791SEric Blake if (offset < 0) { 458dc162c8eSFam Zheng bdrv_set_dirty_iter(s->dbi, 0); 459f798184cSEric Blake offset = bdrv_dirty_iter_next(s->dbi); 4609a46dba7SEric Blake trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 461fb2ef791SEric Blake assert(offset >= 0); 462e5b43573SFam Zheng } 463b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 464e5b43573SFam Zheng 4651181e19aSMax Reitz mirror_wait_on_conflicts(NULL, s, offset, 1); 4669c83625bSMax Reitz 467da01ff7fSKevin Wolf job_pause_point(&s->common.job); 468565ac01fSStefan Hajnoczi 469e5b43573SFam Zheng /* Find the number of consective dirty chunks following the first dirty 470e5b43573SFam Zheng * one, and wait for in flight requests in them. */ 471b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_lock(s->dirty_bitmap); 472fb2ef791SEric Blake while (nb_chunks * s->granularity < s->buf_size) { 473dc162c8eSFam Zheng int64_t next_dirty; 474fb2ef791SEric Blake int64_t next_offset = offset + nb_chunks * s->granularity; 475fb2ef791SEric Blake int64_t next_chunk = next_offset / s->granularity; 476fb2ef791SEric Blake if (next_offset >= s->bdev_length || 4773b5d4df0SEric Blake !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { 478e5b43573SFam Zheng break; 479e5b43573SFam Zheng } 480e5b43573SFam Zheng if (test_bit(next_chunk, s->in_flight_bitmap)) { 481e5b43573SFam Zheng break; 482e5b43573SFam Zheng } 4839c83625bSMax Reitz 484f798184cSEric Blake next_dirty = bdrv_dirty_iter_next(s->dbi); 485fb2ef791SEric Blake if (next_dirty > next_offset || next_dirty < 0) { 486f27a2742SMax Reitz /* The bitmap iterator's cache is stale, refresh it */ 487715a74d8SEric Blake bdrv_set_dirty_iter(s->dbi, next_offset); 488f798184cSEric Blake next_dirty = bdrv_dirty_iter_next(s->dbi); 489f27a2742SMax Reitz } 490fb2ef791SEric Blake assert(next_dirty == next_offset); 491e5b43573SFam Zheng nb_chunks++; 492e5b43573SFam Zheng } 493e5b43573SFam Zheng 494e5b43573SFam Zheng /* Clear dirty bits before querying the block status, because 49531826642SEric Blake * calling bdrv_block_status_above could yield - if some blocks are 496e5b43573SFam Zheng * marked dirty in this window, we need to know. 497e5b43573SFam Zheng */ 498e0d7f73eSEric Blake bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, 499e0d7f73eSEric Blake nb_chunks * s->granularity); 500b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 501b64bd51eSPaolo Bonzini 5021181e19aSMax Reitz /* Before claiming an area in the in-flight bitmap, we have to 5031181e19aSMax Reitz * create a MirrorOp for it so that conflicting requests can wait 5041181e19aSMax Reitz * for it. mirror_perform() will create the real MirrorOps later, 5051181e19aSMax Reitz * for now we just create a pseudo operation that will wake up all 5061181e19aSMax Reitz * conflicting requests once all real operations have been 5071181e19aSMax Reitz * launched. */ 5081181e19aSMax Reitz pseudo_op = g_new(MirrorOp, 1); 5091181e19aSMax Reitz *pseudo_op = (MirrorOp){ 5101181e19aSMax Reitz .offset = offset, 5111181e19aSMax Reitz .bytes = nb_chunks * s->granularity, 5121181e19aSMax Reitz .is_pseudo_op = true, 5131181e19aSMax Reitz }; 5141181e19aSMax Reitz qemu_co_queue_init(&pseudo_op->waiting_requests); 5151181e19aSMax Reitz QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); 5161181e19aSMax Reitz 517fb2ef791SEric Blake bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); 518fb2ef791SEric Blake while (nb_chunks > 0 && offset < s->bdev_length) { 51931826642SEric Blake int ret; 5207cfd5275SEric Blake int64_t io_bytes; 521f3e4ce4aSEric Blake int64_t io_bytes_acct; 5224295c5fcSMax Reitz MirrorMethod mirror_method = MIRROR_METHOD_COPY; 523e5b43573SFam Zheng 524fb2ef791SEric Blake assert(!(offset % s->granularity)); 52531826642SEric Blake ret = bdrv_block_status_above(source, NULL, offset, 52631826642SEric Blake nb_chunks * s->granularity, 52731826642SEric Blake &io_bytes, NULL, NULL); 528e5b43573SFam Zheng if (ret < 0) { 529fb2ef791SEric Blake io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); 5300965a41eSVladimir Sementsov-Ogievskiy } else if (ret & BDRV_BLOCK_DATA) { 531fb2ef791SEric Blake io_bytes = MIN(io_bytes, max_io_bytes); 532e5b43573SFam Zheng } 533e5b43573SFam Zheng 534fb2ef791SEric Blake io_bytes -= io_bytes % s->granularity; 535fb2ef791SEric Blake if (io_bytes < s->granularity) { 536fb2ef791SEric Blake io_bytes = s->granularity; 537e5b43573SFam Zheng } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { 538fb2ef791SEric Blake int64_t target_offset; 5397cfd5275SEric Blake int64_t target_bytes; 540fb2ef791SEric Blake bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, 541fb2ef791SEric Blake &target_offset, &target_bytes); 542fb2ef791SEric Blake if (target_offset == offset && 543fb2ef791SEric Blake target_bytes == io_bytes) { 544e5b43573SFam Zheng mirror_method = ret & BDRV_BLOCK_ZERO ? 545e5b43573SFam Zheng MIRROR_METHOD_ZERO : 546e5b43573SFam Zheng MIRROR_METHOD_DISCARD; 547e5b43573SFam Zheng } 548e5b43573SFam Zheng } 549e5b43573SFam Zheng 550cf56a3c6SDenis V. Lunev while (s->in_flight >= MAX_IN_FLIGHT) { 551fb2ef791SEric Blake trace_mirror_yield_in_flight(s, offset, s->in_flight); 5521181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 553cf56a3c6SDenis V. Lunev } 554cf56a3c6SDenis V. Lunev 555dbaa7b57SVladimir Sementsov-Ogievskiy if (s->ret < 0) { 5561181e19aSMax Reitz ret = 0; 5571181e19aSMax Reitz goto fail; 558dbaa7b57SVladimir Sementsov-Ogievskiy } 559dbaa7b57SVladimir Sementsov-Ogievskiy 560fb2ef791SEric Blake io_bytes = mirror_clip_bytes(s, offset, io_bytes); 5614295c5fcSMax Reitz io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); 5624295c5fcSMax Reitz if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { 563f3e4ce4aSEric Blake io_bytes_acct = 0; 5644b5004d9SDenis V. Lunev } else { 565fb2ef791SEric Blake io_bytes_acct = io_bytes; 5664b5004d9SDenis V. Lunev } 567fb2ef791SEric Blake assert(io_bytes); 568fb2ef791SEric Blake offset += io_bytes; 569fb2ef791SEric Blake nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); 570dee81d51SKevin Wolf delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct); 571dcfb3bebSFam Zheng } 5721181e19aSMax Reitz 5731181e19aSMax Reitz ret = delay_ns; 5741181e19aSMax Reitz fail: 5751181e19aSMax Reitz QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); 5761181e19aSMax Reitz qemu_co_queue_restart_all(&pseudo_op->waiting_requests); 5771181e19aSMax Reitz g_free(pseudo_op); 5781181e19aSMax Reitz 5791181e19aSMax Reitz return ret; 580893f7ebaSPaolo Bonzini } 581b952b558SPaolo Bonzini 582402a4741SPaolo Bonzini static void mirror_free_init(MirrorBlockJob *s) 583402a4741SPaolo Bonzini { 584402a4741SPaolo Bonzini int granularity = s->granularity; 585402a4741SPaolo Bonzini size_t buf_size = s->buf_size; 586402a4741SPaolo Bonzini uint8_t *buf = s->buf; 587402a4741SPaolo Bonzini 588402a4741SPaolo Bonzini assert(s->buf_free_count == 0); 589402a4741SPaolo Bonzini QSIMPLEQ_INIT(&s->buf_free); 590402a4741SPaolo Bonzini while (buf_size != 0) { 591402a4741SPaolo Bonzini MirrorBuffer *cur = (MirrorBuffer *)buf; 592402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 593402a4741SPaolo Bonzini s->buf_free_count++; 594402a4741SPaolo Bonzini buf_size -= granularity; 595402a4741SPaolo Bonzini buf += granularity; 596402a4741SPaolo Bonzini } 597402a4741SPaolo Bonzini } 598402a4741SPaolo Bonzini 599bae8196dSPaolo Bonzini /* This is also used for the .pause callback. There is no matching 600bae8196dSPaolo Bonzini * mirror_resume() because mirror_run() will begin iterating again 601bae8196dSPaolo Bonzini * when the job is resumed. 602bae8196dSPaolo Bonzini */ 603537c3d4fSStefan Hajnoczi static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s) 604bd48bde8SPaolo Bonzini { 605bd48bde8SPaolo Bonzini while (s->in_flight > 0) { 6061181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 607bd48bde8SPaolo Bonzini } 608893f7ebaSPaolo Bonzini } 609893f7ebaSPaolo Bonzini 610737efc1eSJohn Snow /** 611737efc1eSJohn Snow * mirror_exit_common: handle both abort() and prepare() cases. 612737efc1eSJohn Snow * for .prepare, returns 0 on success and -errno on failure. 613737efc1eSJohn Snow * for .abort cases, denoted by abort = true, MUST return 0. 614737efc1eSJohn Snow */ 615737efc1eSJohn Snow static int mirror_exit_common(Job *job) 6165a7e7a0bSStefan Hajnoczi { 6171908a559SKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 6181908a559SKevin Wolf BlockJob *bjob = &s->common; 619429076e8SMax Reitz MirrorBDSOpaque *bs_opaque = s->mirror_top_bs->opaque; 6205a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context = NULL; 621138f9fffSMax Reitz BlockDriverState *src = s->mirror_top_bs->backing->bs; 622e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 6234ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs = s->mirror_top_bs; 62412fa4af6SKevin Wolf Error *local_err = NULL; 625737efc1eSJohn Snow bool abort = job->ret < 0; 626737efc1eSJohn Snow int ret = 0; 627737efc1eSJohn Snow 628737efc1eSJohn Snow if (s->prepared) { 629737efc1eSJohn Snow return 0; 630737efc1eSJohn Snow } 631737efc1eSJohn Snow s->prepared = true; 6323f09bfbcSKevin Wolf 6332119882cSPaolo Bonzini bdrv_release_dirty_bitmap(src, s->dirty_bitmap); 6342119882cSPaolo Bonzini 6357b508f6bSJohn Snow /* Make sure that the source BDS doesn't go away during bdrv_replace_node, 6367b508f6bSJohn Snow * before we can call bdrv_drained_end */ 6373f09bfbcSKevin Wolf bdrv_ref(src); 6384ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 6397d9fcb39SKevin Wolf bdrv_ref(target_bs); 6407d9fcb39SKevin Wolf 6417d9fcb39SKevin Wolf /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before 6427d9fcb39SKevin Wolf * inserting target_bs at s->to_replace, where we might not be able to get 64363c8ef28SKevin Wolf * these permissions. 64463c8ef28SKevin Wolf * 64563c8ef28SKevin Wolf * Note that blk_unref() alone doesn't necessarily drop permissions because 64663c8ef28SKevin Wolf * we might be running nested inside mirror_drain(), which takes an extra 64763c8ef28SKevin Wolf * reference, so use an explicit blk_set_perm() first. */ 64863c8ef28SKevin Wolf blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); 6497d9fcb39SKevin Wolf blk_unref(s->target); 6507d9fcb39SKevin Wolf s->target = NULL; 6514ef85a9cSKevin Wolf 6524ef85a9cSKevin Wolf /* We don't access the source any more. Dropping any WRITE/RESIZE is 6534ef85a9cSKevin Wolf * required before it could become a backing file of target_bs. */ 6544ef85a9cSKevin Wolf bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 6554ef85a9cSKevin Wolf &error_abort); 656737efc1eSJohn Snow if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { 6574ef85a9cSKevin Wolf BlockDriverState *backing = s->is_none_mode ? src : s->base; 6584ef85a9cSKevin Wolf if (backing_bs(target_bs) != backing) { 65912fa4af6SKevin Wolf bdrv_set_backing_hd(target_bs, backing, &local_err); 66012fa4af6SKevin Wolf if (local_err) { 66112fa4af6SKevin Wolf error_report_err(local_err); 6627b508f6bSJohn Snow ret = -EPERM; 66312fa4af6SKevin Wolf } 6644ef85a9cSKevin Wolf } 6654ef85a9cSKevin Wolf } 6665a7e7a0bSStefan Hajnoczi 6675a7e7a0bSStefan Hajnoczi if (s->to_replace) { 6685a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 6695a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 6705a7e7a0bSStefan Hajnoczi } 6715a7e7a0bSStefan Hajnoczi 672737efc1eSJohn Snow if (s->should_complete && !abort) { 673737efc1eSJohn Snow BlockDriverState *to_replace = s->to_replace ?: src; 6741ba79388SAlberto Garcia bool ro = bdrv_is_read_only(to_replace); 67540365552SKevin Wolf 6761ba79388SAlberto Garcia if (ro != bdrv_is_read_only(target_bs)) { 6771ba79388SAlberto Garcia bdrv_reopen_set_read_only(target_bs, ro, NULL); 6785a7e7a0bSStefan Hajnoczi } 679b8804815SKevin Wolf 680b8804815SKevin Wolf /* The mirror job has no requests in flight any more, but we need to 681b8804815SKevin Wolf * drain potential other users of the BDS before changing the graph. */ 682e253f4b8SKevin Wolf bdrv_drained_begin(target_bs); 6835fe31c25SKevin Wolf bdrv_replace_node(to_replace, target_bs, &local_err); 684e253f4b8SKevin Wolf bdrv_drained_end(target_bs); 6855fe31c25SKevin Wolf if (local_err) { 6865fe31c25SKevin Wolf error_report_err(local_err); 6877b508f6bSJohn Snow ret = -EPERM; 6885fe31c25SKevin Wolf } 6895a7e7a0bSStefan Hajnoczi } 6905a7e7a0bSStefan Hajnoczi if (s->to_replace) { 6915a7e7a0bSStefan Hajnoczi bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 6925a7e7a0bSStefan Hajnoczi error_free(s->replace_blocker); 6935a7e7a0bSStefan Hajnoczi bdrv_unref(s->to_replace); 6945a7e7a0bSStefan Hajnoczi } 6955a7e7a0bSStefan Hajnoczi if (replace_aio_context) { 6965a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 6975a7e7a0bSStefan Hajnoczi } 6985a7e7a0bSStefan Hajnoczi g_free(s->replaces); 6997d9fcb39SKevin Wolf bdrv_unref(target_bs); 7004ef85a9cSKevin Wolf 7014ef85a9cSKevin Wolf /* Remove the mirror filter driver from the graph. Before this, get rid of 7024ef85a9cSKevin Wolf * the blockers on the intermediate nodes so that the resulting state is 7030bf74767SKevin Wolf * valid. Also give up permissions on mirror_top_bs->backing, which might 7040bf74767SKevin Wolf * block the removal. */ 7051908a559SKevin Wolf block_job_remove_all_bdrv(bjob); 706c1cef672SFam Zheng bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 707c1cef672SFam Zheng &error_abort); 7085fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 7094ef85a9cSKevin Wolf 7104ef85a9cSKevin Wolf /* We just changed the BDS the job BB refers to (with either or both of the 7115fe31c25SKevin Wolf * bdrv_replace_node() calls), so switch the BB back so the cleanup does 7125fe31c25SKevin Wolf * the right thing. We don't need any permissions any more now. */ 7131908a559SKevin Wolf blk_remove_bs(bjob->blk); 7141908a559SKevin Wolf blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); 7151908a559SKevin Wolf blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort); 7164ef85a9cSKevin Wolf 717429076e8SMax Reitz bs_opaque->job = NULL; 7184ef85a9cSKevin Wolf 719176c3699SFam Zheng bdrv_drained_end(src); 7204ef85a9cSKevin Wolf bdrv_unref(mirror_top_bs); 7213f09bfbcSKevin Wolf bdrv_unref(src); 7227b508f6bSJohn Snow 723737efc1eSJohn Snow return ret; 724737efc1eSJohn Snow } 725737efc1eSJohn Snow 726737efc1eSJohn Snow static int mirror_prepare(Job *job) 727737efc1eSJohn Snow { 728737efc1eSJohn Snow return mirror_exit_common(job); 729737efc1eSJohn Snow } 730737efc1eSJohn Snow 731737efc1eSJohn Snow static void mirror_abort(Job *job) 732737efc1eSJohn Snow { 733737efc1eSJohn Snow int ret = mirror_exit_common(job); 734737efc1eSJohn Snow assert(ret == 0); 7355a7e7a0bSStefan Hajnoczi } 7365a7e7a0bSStefan Hajnoczi 737537c3d4fSStefan Hajnoczi static void coroutine_fn mirror_throttle(MirrorBlockJob *s) 73849efb1f5SDenis V. Lunev { 73949efb1f5SDenis V. Lunev int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 74049efb1f5SDenis V. Lunev 74118bb6928SKevin Wolf if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { 74249efb1f5SDenis V. Lunev s->last_pause_ns = now; 7435d43e86eSKevin Wolf job_sleep_ns(&s->common.job, 0); 74449efb1f5SDenis V. Lunev } else { 745da01ff7fSKevin Wolf job_pause_point(&s->common.job); 74649efb1f5SDenis V. Lunev } 74749efb1f5SDenis V. Lunev } 74849efb1f5SDenis V. Lunev 749c0b363adSDenis V. Lunev static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) 750c0b363adSDenis V. Lunev { 75123ca459aSEric Blake int64_t offset; 752c0b363adSDenis V. Lunev BlockDriverState *base = s->base; 753138f9fffSMax Reitz BlockDriverState *bs = s->mirror_top_bs->backing->bs; 754c0b363adSDenis V. Lunev BlockDriverState *target_bs = blk_bs(s->target); 75523ca459aSEric Blake int ret; 75651b0a488SEric Blake int64_t count; 757c0b363adSDenis V. Lunev 758b7d5062cSDenis V. Lunev if (base == NULL && !bdrv_has_zero_init(target_bs)) { 759c7c2769cSDenis V. Lunev if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { 760e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); 761b7d5062cSDenis V. Lunev return 0; 762b7d5062cSDenis V. Lunev } 763b7d5062cSDenis V. Lunev 76490ab48ebSAnton Nefedov s->initial_zeroing_ongoing = true; 76523ca459aSEric Blake for (offset = 0; offset < s->bdev_length; ) { 76623ca459aSEric Blake int bytes = MIN(s->bdev_length - offset, 76723ca459aSEric Blake QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 768c7c2769cSDenis V. Lunev 769c7c2769cSDenis V. Lunev mirror_throttle(s); 770c7c2769cSDenis V. Lunev 771daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job)) { 77290ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 773c7c2769cSDenis V. Lunev return 0; 774c7c2769cSDenis V. Lunev } 775c7c2769cSDenis V. Lunev 776c7c2769cSDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT) { 77767adf4b3SEric Blake trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, 77867adf4b3SEric Blake s->in_flight); 7791181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 780c7c2769cSDenis V. Lunev continue; 781c7c2769cSDenis V. Lunev } 782c7c2769cSDenis V. Lunev 7834295c5fcSMax Reitz mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); 78423ca459aSEric Blake offset += bytes; 785c7c2769cSDenis V. Lunev } 786c7c2769cSDenis V. Lunev 787bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 78890ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 789c7c2769cSDenis V. Lunev } 790c7c2769cSDenis V. Lunev 791c0b363adSDenis V. Lunev /* First part, loop on the sectors and initialize the dirty bitmap. */ 79223ca459aSEric Blake for (offset = 0; offset < s->bdev_length; ) { 793c0b363adSDenis V. Lunev /* Just to make sure we are not exceeding int limit. */ 79423ca459aSEric Blake int bytes = MIN(s->bdev_length - offset, 79523ca459aSEric Blake QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 796c0b363adSDenis V. Lunev 797c0b363adSDenis V. Lunev mirror_throttle(s); 798c0b363adSDenis V. Lunev 799daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job)) { 800c0b363adSDenis V. Lunev return 0; 801c0b363adSDenis V. Lunev } 802c0b363adSDenis V. Lunev 80323ca459aSEric Blake ret = bdrv_is_allocated_above(bs, base, offset, bytes, &count); 804c0b363adSDenis V. Lunev if (ret < 0) { 805c0b363adSDenis V. Lunev return ret; 806c0b363adSDenis V. Lunev } 807c0b363adSDenis V. Lunev 80823ca459aSEric Blake assert(count); 809b7d5062cSDenis V. Lunev if (ret == 1) { 81023ca459aSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); 811c0b363adSDenis V. Lunev } 81223ca459aSEric Blake offset += count; 813c0b363adSDenis V. Lunev } 814c0b363adSDenis V. Lunev return 0; 815c0b363adSDenis V. Lunev } 816c0b363adSDenis V. Lunev 817bdffb31dSPaolo Bonzini /* Called when going out of the streaming phase to flush the bulk of the 818bdffb31dSPaolo Bonzini * data to the medium, or just before completing. 819bdffb31dSPaolo Bonzini */ 820bdffb31dSPaolo Bonzini static int mirror_flush(MirrorBlockJob *s) 821bdffb31dSPaolo Bonzini { 822bdffb31dSPaolo Bonzini int ret = blk_flush(s->target); 823bdffb31dSPaolo Bonzini if (ret < 0) { 824bdffb31dSPaolo Bonzini if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { 825bdffb31dSPaolo Bonzini s->ret = ret; 826bdffb31dSPaolo Bonzini } 827bdffb31dSPaolo Bonzini } 828bdffb31dSPaolo Bonzini return ret; 829bdffb31dSPaolo Bonzini } 830bdffb31dSPaolo Bonzini 831f67432a2SJohn Snow static int coroutine_fn mirror_run(Job *job, Error **errp) 832893f7ebaSPaolo Bonzini { 833f67432a2SJohn Snow MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 834138f9fffSMax Reitz BlockDriverState *bs = s->mirror_top_bs->backing->bs; 835e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 8369a0cec66SPaolo Bonzini bool need_drain = true; 837c0b363adSDenis V. Lunev int64_t length; 838b812f671SPaolo Bonzini BlockDriverInfo bdi; 8391d33936eSJeff Cody char backing_filename[2]; /* we only need 2 characters because we are only 8401d33936eSJeff Cody checking for a NULL string */ 841893f7ebaSPaolo Bonzini int ret = 0; 842893f7ebaSPaolo Bonzini 843daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job)) { 844893f7ebaSPaolo Bonzini goto immediate_exit; 845893f7ebaSPaolo Bonzini } 846893f7ebaSPaolo Bonzini 847b21c7652SMax Reitz s->bdev_length = bdrv_getlength(bs); 848b21c7652SMax Reitz if (s->bdev_length < 0) { 849b21c7652SMax Reitz ret = s->bdev_length; 850373df5b1SFam Zheng goto immediate_exit; 851becc347eSKevin Wolf } 852becc347eSKevin Wolf 853becc347eSKevin Wolf /* Active commit must resize the base image if its size differs from the 854becc347eSKevin Wolf * active layer. */ 855becc347eSKevin Wolf if (s->base == blk_bs(s->target)) { 856becc347eSKevin Wolf int64_t base_length; 857becc347eSKevin Wolf 858becc347eSKevin Wolf base_length = blk_getlength(s->target); 859becc347eSKevin Wolf if (base_length < 0) { 860becc347eSKevin Wolf ret = base_length; 861becc347eSKevin Wolf goto immediate_exit; 862becc347eSKevin Wolf } 863becc347eSKevin Wolf 864becc347eSKevin Wolf if (s->bdev_length > base_length) { 8653a691c50SMax Reitz ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, 8663a691c50SMax Reitz NULL); 867becc347eSKevin Wolf if (ret < 0) { 868becc347eSKevin Wolf goto immediate_exit; 869becc347eSKevin Wolf } 870becc347eSKevin Wolf } 871becc347eSKevin Wolf } 872becc347eSKevin Wolf 873becc347eSKevin Wolf if (s->bdev_length == 0) { 8742e1795b5SKevin Wolf /* Transition to the READY state and wait for complete. */ 8752e1795b5SKevin Wolf job_transition_to_ready(&s->common.job); 8769e48b025SFam Zheng s->synced = true; 877d06107adSMax Reitz s->actively_synced = true; 878daa7f2f9SKevin Wolf while (!job_is_cancelled(&s->common.job) && !s->should_complete) { 879198c49ccSKevin Wolf job_yield(&s->common.job); 8809e48b025SFam Zheng } 881daa7f2f9SKevin Wolf s->common.job.cancelled = false; 8829e48b025SFam Zheng goto immediate_exit; 883893f7ebaSPaolo Bonzini } 884893f7ebaSPaolo Bonzini 885b21c7652SMax Reitz length = DIV_ROUND_UP(s->bdev_length, s->granularity); 886402a4741SPaolo Bonzini s->in_flight_bitmap = bitmap_new(length); 887402a4741SPaolo Bonzini 888b812f671SPaolo Bonzini /* If we have no backing file yet in the destination, we cannot let 889b812f671SPaolo Bonzini * the destination do COW. Instead, we copy sectors around the 890b812f671SPaolo Bonzini * dirty data if needed. We need a bitmap to do that. 891b812f671SPaolo Bonzini */ 892e253f4b8SKevin Wolf bdrv_get_backing_filename(target_bs, backing_filename, 893b812f671SPaolo Bonzini sizeof(backing_filename)); 894e253f4b8SKevin Wolf if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { 895b436982fSEric Blake s->target_cluster_size = bdi.cluster_size; 896b436982fSEric Blake } else { 897b436982fSEric Blake s->target_cluster_size = BDRV_SECTOR_SIZE; 898c3cc95bdSFam Zheng } 899b436982fSEric Blake if (backing_filename[0] && !target_bs->backing && 900b436982fSEric Blake s->granularity < s->target_cluster_size) { 901b436982fSEric Blake s->buf_size = MAX(s->buf_size, s->target_cluster_size); 902b812f671SPaolo Bonzini s->cow_bitmap = bitmap_new(length); 903b812f671SPaolo Bonzini } 904e253f4b8SKevin Wolf s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); 905b812f671SPaolo Bonzini 9067504edf4SKevin Wolf s->buf = qemu_try_blockalign(bs, s->buf_size); 9077504edf4SKevin Wolf if (s->buf == NULL) { 9087504edf4SKevin Wolf ret = -ENOMEM; 9097504edf4SKevin Wolf goto immediate_exit; 9107504edf4SKevin Wolf } 9117504edf4SKevin Wolf 912402a4741SPaolo Bonzini mirror_free_init(s); 913893f7ebaSPaolo Bonzini 91449efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 91503544a6eSFam Zheng if (!s->is_none_mode) { 916c0b363adSDenis V. Lunev ret = mirror_dirty_init(s); 917daa7f2f9SKevin Wolf if (ret < 0 || job_is_cancelled(&s->common.job)) { 9184c0cbd6fSFam Zheng goto immediate_exit; 9194c0cbd6fSFam Zheng } 920893f7ebaSPaolo Bonzini } 921893f7ebaSPaolo Bonzini 922dc162c8eSFam Zheng assert(!s->dbi); 923715a74d8SEric Blake s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); 924893f7ebaSPaolo Bonzini for (;;) { 925cc8c9d6cSPaolo Bonzini uint64_t delay_ns = 0; 92649efb1f5SDenis V. Lunev int64_t cnt, delta; 927893f7ebaSPaolo Bonzini bool should_complete; 928893f7ebaSPaolo Bonzini 929d06107adSMax Reitz /* Do not start passive operations while there are active 930d06107adSMax Reitz * writes in progress */ 931d06107adSMax Reitz while (s->in_active_write_counter) { 932d06107adSMax Reitz mirror_wait_for_any_operation(s, true); 933d06107adSMax Reitz } 934d06107adSMax Reitz 935bd48bde8SPaolo Bonzini if (s->ret < 0) { 936bd48bde8SPaolo Bonzini ret = s->ret; 937893f7ebaSPaolo Bonzini goto immediate_exit; 938893f7ebaSPaolo Bonzini } 939bd48bde8SPaolo Bonzini 940da01ff7fSKevin Wolf job_pause_point(&s->common.job); 941565ac01fSStefan Hajnoczi 94220dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 94305df8a6aSKevin Wolf /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is 94405df8a6aSKevin Wolf * the number of bytes currently being processed; together those are 94505df8a6aSKevin Wolf * the current remaining operation length */ 94630a5c887SKevin Wolf job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt); 947bd48bde8SPaolo Bonzini 948bd48bde8SPaolo Bonzini /* Note that even when no rate limit is applied we need to yield 949a7282330SFam Zheng * periodically with no pending I/O so that bdrv_drain_all() returns. 95018bb6928SKevin Wolf * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is 95118bb6928SKevin Wolf * an error, or when the source is clean, whichever comes first. */ 95249efb1f5SDenis V. Lunev delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; 95318bb6928SKevin Wolf if (delta < BLOCK_JOB_SLICE_TIME && 954bd48bde8SPaolo Bonzini s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 955cf56a3c6SDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || 956402a4741SPaolo Bonzini (cnt == 0 && s->in_flight > 0)) { 9579a46dba7SEric Blake trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); 9581181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 959bd48bde8SPaolo Bonzini continue; 960bd48bde8SPaolo Bonzini } else if (cnt != 0) { 961cc8c9d6cSPaolo Bonzini delay_ns = mirror_iteration(s); 962893f7ebaSPaolo Bonzini } 963cc8c9d6cSPaolo Bonzini } 964893f7ebaSPaolo Bonzini 965893f7ebaSPaolo Bonzini should_complete = false; 966bd48bde8SPaolo Bonzini if (s->in_flight == 0 && cnt == 0) { 967893f7ebaSPaolo Bonzini trace_mirror_before_flush(s); 968bdffb31dSPaolo Bonzini if (!s->synced) { 969bdffb31dSPaolo Bonzini if (mirror_flush(s) < 0) { 970bdffb31dSPaolo Bonzini /* Go check s->ret. */ 971bdffb31dSPaolo Bonzini continue; 972893f7ebaSPaolo Bonzini } 973893f7ebaSPaolo Bonzini /* We're out of the streaming phase. From now on, if the job 974893f7ebaSPaolo Bonzini * is cancelled we will actually complete all pending I/O and 975893f7ebaSPaolo Bonzini * report completion. This way, block-job-cancel will leave 976893f7ebaSPaolo Bonzini * the target in a consistent state. 977893f7ebaSPaolo Bonzini */ 9782e1795b5SKevin Wolf job_transition_to_ready(&s->common.job); 979d63ffd87SPaolo Bonzini s->synced = true; 980d06107adSMax Reitz if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) { 981d06107adSMax Reitz s->actively_synced = true; 982d06107adSMax Reitz } 983d63ffd87SPaolo Bonzini } 984d63ffd87SPaolo Bonzini 985d63ffd87SPaolo Bonzini should_complete = s->should_complete || 986daa7f2f9SKevin Wolf job_is_cancelled(&s->common.job); 98720dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 988893f7ebaSPaolo Bonzini } 989893f7ebaSPaolo Bonzini 990893f7ebaSPaolo Bonzini if (cnt == 0 && should_complete) { 991893f7ebaSPaolo Bonzini /* The dirty bitmap is not updated while operations are pending. 992893f7ebaSPaolo Bonzini * If we're about to exit, wait for pending operations before 993893f7ebaSPaolo Bonzini * calling bdrv_get_dirty_count(bs), or we may exit while the 994893f7ebaSPaolo Bonzini * source has dirty data to copy! 995893f7ebaSPaolo Bonzini * 996893f7ebaSPaolo Bonzini * Note that I/O can be submitted by the guest while 9979a0cec66SPaolo Bonzini * mirror_populate runs, so pause it now. Before deciding 9989a0cec66SPaolo Bonzini * whether to switch to target check one last time if I/O has 9999a0cec66SPaolo Bonzini * come in the meanwhile, and if not flush the data to disk. 1000893f7ebaSPaolo Bonzini */ 10019a46dba7SEric Blake trace_mirror_before_drain(s, cnt); 10029a0cec66SPaolo Bonzini 10039a0cec66SPaolo Bonzini bdrv_drained_begin(bs); 100420dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 1005bdffb31dSPaolo Bonzini if (cnt > 0 || mirror_flush(s) < 0) { 10069a0cec66SPaolo Bonzini bdrv_drained_end(bs); 10079a0cec66SPaolo Bonzini continue; 10089a0cec66SPaolo Bonzini } 10099a0cec66SPaolo Bonzini 10109a0cec66SPaolo Bonzini /* The two disks are in sync. Exit and report successful 10119a0cec66SPaolo Bonzini * completion. 10129a0cec66SPaolo Bonzini */ 10139a0cec66SPaolo Bonzini assert(QLIST_EMPTY(&bs->tracked_requests)); 1014daa7f2f9SKevin Wolf s->common.job.cancelled = false; 10159a0cec66SPaolo Bonzini need_drain = false; 10169a0cec66SPaolo Bonzini break; 1017893f7ebaSPaolo Bonzini } 1018893f7ebaSPaolo Bonzini 1019893f7ebaSPaolo Bonzini ret = 0; 1020ddc4115eSStefan Hajnoczi 1021ddc4115eSStefan Hajnoczi if (s->synced && !should_complete) { 102218bb6928SKevin Wolf delay_ns = (s->in_flight == 0 && 102318bb6928SKevin Wolf cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0); 1024ddc4115eSStefan Hajnoczi } 10259a46dba7SEric Blake trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); 10265d43e86eSKevin Wolf job_sleep_ns(&s->common.job, delay_ns); 1027daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job) && 1028004e95dfSKevin Wolf (!s->synced || s->common.job.force_cancel)) 1029eb36639fSMax Reitz { 1030893f7ebaSPaolo Bonzini break; 1031893f7ebaSPaolo Bonzini } 103249efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1033893f7ebaSPaolo Bonzini } 1034893f7ebaSPaolo Bonzini 1035893f7ebaSPaolo Bonzini immediate_exit: 1036bd48bde8SPaolo Bonzini if (s->in_flight > 0) { 1037bd48bde8SPaolo Bonzini /* We get here only if something went wrong. Either the job failed, 1038bd48bde8SPaolo Bonzini * or it was cancelled prematurely so that we do not guarantee that 1039bd48bde8SPaolo Bonzini * the target is a copy of the source. 1040bd48bde8SPaolo Bonzini */ 1041004e95dfSKevin Wolf assert(ret < 0 || ((s->common.job.force_cancel || !s->synced) && 1042daa7f2f9SKevin Wolf job_is_cancelled(&s->common.job))); 10439a0cec66SPaolo Bonzini assert(need_drain); 1044bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 1045bd48bde8SPaolo Bonzini } 1046bd48bde8SPaolo Bonzini 1047bd48bde8SPaolo Bonzini assert(s->in_flight == 0); 10487191bf31SMarkus Armbruster qemu_vfree(s->buf); 1049b812f671SPaolo Bonzini g_free(s->cow_bitmap); 1050402a4741SPaolo Bonzini g_free(s->in_flight_bitmap); 1051dc162c8eSFam Zheng bdrv_dirty_iter_free(s->dbi); 10525a7e7a0bSStefan Hajnoczi 10539a0cec66SPaolo Bonzini if (need_drain) { 1054e253f4b8SKevin Wolf bdrv_drained_begin(bs); 10559a0cec66SPaolo Bonzini } 1056f67432a2SJohn Snow 1057f67432a2SJohn Snow return ret; 1058893f7ebaSPaolo Bonzini } 1059893f7ebaSPaolo Bonzini 10603453d972SKevin Wolf static void mirror_complete(Job *job, Error **errp) 1061d63ffd87SPaolo Bonzini { 10623453d972SKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 10634ef85a9cSKevin Wolf BlockDriverState *target; 1064d63ffd87SPaolo Bonzini 1065274fcceeSMax Reitz target = blk_bs(s->target); 1066274fcceeSMax Reitz 1067d63ffd87SPaolo Bonzini if (!s->synced) { 10689df229c3SAlberto Garcia error_setg(errp, "The active block job '%s' cannot be completed", 10693453d972SKevin Wolf job->id); 1070d63ffd87SPaolo Bonzini return; 1071d63ffd87SPaolo Bonzini } 1072d63ffd87SPaolo Bonzini 1073274fcceeSMax Reitz if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { 1074274fcceeSMax Reitz int ret; 1075274fcceeSMax Reitz 1076274fcceeSMax Reitz assert(!target->backing); 1077274fcceeSMax Reitz ret = bdrv_open_backing_file(target, NULL, "backing", errp); 1078274fcceeSMax Reitz if (ret < 0) { 1079274fcceeSMax Reitz return; 1080274fcceeSMax Reitz } 1081274fcceeSMax Reitz } 1082274fcceeSMax Reitz 108315d67298SChanglong Xie /* block all operations on to_replace bs */ 108409158f00SBenoît Canet if (s->replaces) { 10855a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context; 10865a7e7a0bSStefan Hajnoczi 1087e12f3784SWen Congyang s->to_replace = bdrv_find_node(s->replaces); 108809158f00SBenoît Canet if (!s->to_replace) { 1089e12f3784SWen Congyang error_setg(errp, "Node name '%s' not found", s->replaces); 109009158f00SBenoît Canet return; 109109158f00SBenoît Canet } 109209158f00SBenoît Canet 10935a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 10945a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 10955a7e7a0bSStefan Hajnoczi 10964ef85a9cSKevin Wolf /* TODO Translate this into permission system. Current definition of 10974ef85a9cSKevin Wolf * GRAPH_MOD would require to request it for the parents; they might 10984ef85a9cSKevin Wolf * not even be BlockDriverStates, however, so a BdrvChild can't address 10994ef85a9cSKevin Wolf * them. May need redefinition of GRAPH_MOD. */ 110009158f00SBenoît Canet error_setg(&s->replace_blocker, 110109158f00SBenoît Canet "block device is in use by block-job-complete"); 110209158f00SBenoît Canet bdrv_op_block_all(s->to_replace, s->replace_blocker); 110309158f00SBenoît Canet bdrv_ref(s->to_replace); 11045a7e7a0bSStefan Hajnoczi 11055a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 110609158f00SBenoît Canet } 110709158f00SBenoît Canet 1108d63ffd87SPaolo Bonzini s->should_complete = true; 11093d70ff53SKevin Wolf job_enter(job); 1110d63ffd87SPaolo Bonzini } 1111d63ffd87SPaolo Bonzini 1112537c3d4fSStefan Hajnoczi static void coroutine_fn mirror_pause(Job *job) 1113565ac01fSStefan Hajnoczi { 1114da01ff7fSKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 1115565ac01fSStefan Hajnoczi 1116bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 1117565ac01fSStefan Hajnoczi } 1118565ac01fSStefan Hajnoczi 111989bd0305SKevin Wolf static bool mirror_drained_poll(BlockJob *job) 112089bd0305SKevin Wolf { 112189bd0305SKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 112289bd0305SKevin Wolf return !!s->in_flight; 112389bd0305SKevin Wolf } 112489bd0305SKevin Wolf 1125565ac01fSStefan Hajnoczi static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) 1126565ac01fSStefan Hajnoczi { 1127565ac01fSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 1128565ac01fSStefan Hajnoczi 1129565ac01fSStefan Hajnoczi blk_set_aio_context(s->target, new_context); 1130565ac01fSStefan Hajnoczi } 1131565ac01fSStefan Hajnoczi 1132bae8196dSPaolo Bonzini static void mirror_drain(BlockJob *job) 1133bae8196dSPaolo Bonzini { 1134bae8196dSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 1135bae8196dSPaolo Bonzini 1136bae8196dSPaolo Bonzini /* Need to keep a reference in case blk_drain triggers execution 1137bae8196dSPaolo Bonzini * of mirror_complete... 1138bae8196dSPaolo Bonzini */ 1139bae8196dSPaolo Bonzini if (s->target) { 1140bae8196dSPaolo Bonzini BlockBackend *target = s->target; 1141bae8196dSPaolo Bonzini blk_ref(target); 1142bae8196dSPaolo Bonzini blk_drain(target); 1143bae8196dSPaolo Bonzini blk_unref(target); 1144bae8196dSPaolo Bonzini } 1145bae8196dSPaolo Bonzini } 1146bae8196dSPaolo Bonzini 11473fc4b10aSFam Zheng static const BlockJobDriver mirror_job_driver = { 114833e9e9bdSKevin Wolf .job_driver = { 1149893f7ebaSPaolo Bonzini .instance_size = sizeof(MirrorBlockJob), 11508e4c8700SKevin Wolf .job_type = JOB_TYPE_MIRROR, 115180fa2c75SKevin Wolf .free = block_job_free, 1152b15de828SKevin Wolf .user_resume = block_job_user_resume, 1153b69f777dSKevin Wolf .drain = block_job_drain, 1154f67432a2SJohn Snow .run = mirror_run, 1155737efc1eSJohn Snow .prepare = mirror_prepare, 1156737efc1eSJohn Snow .abort = mirror_abort, 1157565ac01fSStefan Hajnoczi .pause = mirror_pause, 1158da01ff7fSKevin Wolf .complete = mirror_complete, 11593453d972SKevin Wolf }, 116089bd0305SKevin Wolf .drained_poll = mirror_drained_poll, 1161565ac01fSStefan Hajnoczi .attached_aio_context = mirror_attached_aio_context, 1162bae8196dSPaolo Bonzini .drain = mirror_drain, 1163893f7ebaSPaolo Bonzini }; 1164893f7ebaSPaolo Bonzini 116503544a6eSFam Zheng static const BlockJobDriver commit_active_job_driver = { 116633e9e9bdSKevin Wolf .job_driver = { 116703544a6eSFam Zheng .instance_size = sizeof(MirrorBlockJob), 11688e4c8700SKevin Wolf .job_type = JOB_TYPE_COMMIT, 116980fa2c75SKevin Wolf .free = block_job_free, 1170b15de828SKevin Wolf .user_resume = block_job_user_resume, 1171b69f777dSKevin Wolf .drain = block_job_drain, 1172f67432a2SJohn Snow .run = mirror_run, 1173737efc1eSJohn Snow .prepare = mirror_prepare, 1174737efc1eSJohn Snow .abort = mirror_abort, 1175565ac01fSStefan Hajnoczi .pause = mirror_pause, 1176da01ff7fSKevin Wolf .complete = mirror_complete, 11773453d972SKevin Wolf }, 117889bd0305SKevin Wolf .drained_poll = mirror_drained_poll, 1179565ac01fSStefan Hajnoczi .attached_aio_context = mirror_attached_aio_context, 1180bae8196dSPaolo Bonzini .drain = mirror_drain, 118103544a6eSFam Zheng }; 118203544a6eSFam Zheng 1183537c3d4fSStefan Hajnoczi static void coroutine_fn 1184537c3d4fSStefan Hajnoczi do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, 1185d06107adSMax Reitz uint64_t offset, uint64_t bytes, 1186d06107adSMax Reitz QEMUIOVector *qiov, int flags) 1187d06107adSMax Reitz { 1188d06107adSMax Reitz QEMUIOVector target_qiov; 1189*1eaf1b0fSVladimir Sementsov-Ogievskiy uint64_t dirty_offset = offset; 1190*1eaf1b0fSVladimir Sementsov-Ogievskiy uint64_t dirty_bytes; 1191d06107adSMax Reitz 1192d06107adSMax Reitz if (qiov) { 1193d06107adSMax Reitz qemu_iovec_init(&target_qiov, qiov->niov); 1194d06107adSMax Reitz } 1195d06107adSMax Reitz 1196d06107adSMax Reitz while (true) { 1197d06107adSMax Reitz bool valid_area; 1198d06107adSMax Reitz int ret; 1199d06107adSMax Reitz 1200d06107adSMax Reitz bdrv_dirty_bitmap_lock(job->dirty_bitmap); 1201*1eaf1b0fSVladimir Sementsov-Ogievskiy dirty_bytes = MIN(offset + bytes - dirty_offset, INT_MAX); 1202*1eaf1b0fSVladimir Sementsov-Ogievskiy valid_area = bdrv_dirty_bitmap_next_dirty_area(job->dirty_bitmap, 1203*1eaf1b0fSVladimir Sementsov-Ogievskiy &dirty_offset, 1204*1eaf1b0fSVladimir Sementsov-Ogievskiy &dirty_bytes); 1205d06107adSMax Reitz if (!valid_area) { 1206d06107adSMax Reitz bdrv_dirty_bitmap_unlock(job->dirty_bitmap); 1207d06107adSMax Reitz break; 1208d06107adSMax Reitz } 1209d06107adSMax Reitz 1210d06107adSMax Reitz bdrv_reset_dirty_bitmap_locked(job->dirty_bitmap, 1211d06107adSMax Reitz dirty_offset, dirty_bytes); 1212d06107adSMax Reitz bdrv_dirty_bitmap_unlock(job->dirty_bitmap); 1213d06107adSMax Reitz 1214d06107adSMax Reitz job_progress_increase_remaining(&job->common.job, dirty_bytes); 1215d06107adSMax Reitz 1216d06107adSMax Reitz assert(dirty_offset - offset <= SIZE_MAX); 1217d06107adSMax Reitz if (qiov) { 1218d06107adSMax Reitz qemu_iovec_reset(&target_qiov); 1219d06107adSMax Reitz qemu_iovec_concat(&target_qiov, qiov, 1220d06107adSMax Reitz dirty_offset - offset, dirty_bytes); 1221d06107adSMax Reitz } 1222d06107adSMax Reitz 1223d06107adSMax Reitz switch (method) { 1224d06107adSMax Reitz case MIRROR_METHOD_COPY: 1225d06107adSMax Reitz ret = blk_co_pwritev(job->target, dirty_offset, dirty_bytes, 1226d06107adSMax Reitz qiov ? &target_qiov : NULL, flags); 1227d06107adSMax Reitz break; 1228d06107adSMax Reitz 1229d06107adSMax Reitz case MIRROR_METHOD_ZERO: 1230d06107adSMax Reitz assert(!qiov); 1231d06107adSMax Reitz ret = blk_co_pwrite_zeroes(job->target, dirty_offset, dirty_bytes, 1232d06107adSMax Reitz flags); 1233d06107adSMax Reitz break; 1234d06107adSMax Reitz 1235d06107adSMax Reitz case MIRROR_METHOD_DISCARD: 1236d06107adSMax Reitz assert(!qiov); 1237d06107adSMax Reitz ret = blk_co_pdiscard(job->target, dirty_offset, dirty_bytes); 1238d06107adSMax Reitz break; 1239d06107adSMax Reitz 1240d06107adSMax Reitz default: 1241d06107adSMax Reitz abort(); 1242d06107adSMax Reitz } 1243d06107adSMax Reitz 1244d06107adSMax Reitz if (ret >= 0) { 1245d06107adSMax Reitz job_progress_update(&job->common.job, dirty_bytes); 1246d06107adSMax Reitz } else { 1247d06107adSMax Reitz BlockErrorAction action; 1248d06107adSMax Reitz 1249d06107adSMax Reitz bdrv_set_dirty_bitmap(job->dirty_bitmap, dirty_offset, dirty_bytes); 1250d06107adSMax Reitz job->actively_synced = false; 1251d06107adSMax Reitz 1252d06107adSMax Reitz action = mirror_error_action(job, false, -ret); 1253d06107adSMax Reitz if (action == BLOCK_ERROR_ACTION_REPORT) { 1254d06107adSMax Reitz if (!job->ret) { 1255d06107adSMax Reitz job->ret = ret; 1256d06107adSMax Reitz } 1257d06107adSMax Reitz break; 1258d06107adSMax Reitz } 1259d06107adSMax Reitz } 1260*1eaf1b0fSVladimir Sementsov-Ogievskiy 1261*1eaf1b0fSVladimir Sementsov-Ogievskiy dirty_offset += dirty_bytes; 1262d06107adSMax Reitz } 1263d06107adSMax Reitz 1264d06107adSMax Reitz if (qiov) { 1265d06107adSMax Reitz qemu_iovec_destroy(&target_qiov); 1266d06107adSMax Reitz } 1267d06107adSMax Reitz } 1268d06107adSMax Reitz 1269d06107adSMax Reitz static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, 1270d06107adSMax Reitz uint64_t offset, 1271d06107adSMax Reitz uint64_t bytes) 1272d06107adSMax Reitz { 1273d06107adSMax Reitz MirrorOp *op; 1274d06107adSMax Reitz uint64_t start_chunk = offset / s->granularity; 1275d06107adSMax Reitz uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); 1276d06107adSMax Reitz 1277d06107adSMax Reitz op = g_new(MirrorOp, 1); 1278d06107adSMax Reitz *op = (MirrorOp){ 1279d06107adSMax Reitz .s = s, 1280d06107adSMax Reitz .offset = offset, 1281d06107adSMax Reitz .bytes = bytes, 1282d06107adSMax Reitz .is_active_write = true, 1283d06107adSMax Reitz }; 1284d06107adSMax Reitz qemu_co_queue_init(&op->waiting_requests); 1285d06107adSMax Reitz QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); 1286d06107adSMax Reitz 1287d06107adSMax Reitz s->in_active_write_counter++; 1288d06107adSMax Reitz 1289d06107adSMax Reitz mirror_wait_on_conflicts(op, s, offset, bytes); 1290d06107adSMax Reitz 1291d06107adSMax Reitz bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); 1292d06107adSMax Reitz 1293d06107adSMax Reitz return op; 1294d06107adSMax Reitz } 1295d06107adSMax Reitz 1296d06107adSMax Reitz static void coroutine_fn active_write_settle(MirrorOp *op) 1297d06107adSMax Reitz { 1298d06107adSMax Reitz uint64_t start_chunk = op->offset / op->s->granularity; 1299d06107adSMax Reitz uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, 1300d06107adSMax Reitz op->s->granularity); 1301d06107adSMax Reitz 1302d06107adSMax Reitz if (!--op->s->in_active_write_counter && op->s->actively_synced) { 1303d06107adSMax Reitz BdrvChild *source = op->s->mirror_top_bs->backing; 1304d06107adSMax Reitz 1305d06107adSMax Reitz if (QLIST_FIRST(&source->bs->parents) == source && 1306d06107adSMax Reitz QLIST_NEXT(source, next_parent) == NULL) 1307d06107adSMax Reitz { 1308d06107adSMax Reitz /* Assert that we are back in sync once all active write 1309d06107adSMax Reitz * operations are settled. 1310d06107adSMax Reitz * Note that we can only assert this if the mirror node 1311d06107adSMax Reitz * is the source node's only parent. */ 1312d06107adSMax Reitz assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); 1313d06107adSMax Reitz } 1314d06107adSMax Reitz } 1315d06107adSMax Reitz bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); 1316d06107adSMax Reitz QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); 1317d06107adSMax Reitz qemu_co_queue_restart_all(&op->waiting_requests); 1318d06107adSMax Reitz g_free(op); 1319d06107adSMax Reitz } 1320d06107adSMax Reitz 13214ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, 13224ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 13234ef85a9cSKevin Wolf { 13244ef85a9cSKevin Wolf return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); 13254ef85a9cSKevin Wolf } 13264ef85a9cSKevin Wolf 1327d06107adSMax Reitz static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs, 1328d06107adSMax Reitz MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, 1329d06107adSMax Reitz int flags) 1330d06107adSMax Reitz { 1331d06107adSMax Reitz MirrorOp *op = NULL; 1332d06107adSMax Reitz MirrorBDSOpaque *s = bs->opaque; 1333d06107adSMax Reitz int ret = 0; 1334d06107adSMax Reitz bool copy_to_target; 1335d06107adSMax Reitz 1336d06107adSMax Reitz copy_to_target = s->job->ret >= 0 && 1337d06107adSMax Reitz s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; 1338d06107adSMax Reitz 1339d06107adSMax Reitz if (copy_to_target) { 1340d06107adSMax Reitz op = active_write_prepare(s->job, offset, bytes); 1341d06107adSMax Reitz } 1342d06107adSMax Reitz 1343d06107adSMax Reitz switch (method) { 1344d06107adSMax Reitz case MIRROR_METHOD_COPY: 1345d06107adSMax Reitz ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); 1346d06107adSMax Reitz break; 1347d06107adSMax Reitz 1348d06107adSMax Reitz case MIRROR_METHOD_ZERO: 1349d06107adSMax Reitz ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); 1350d06107adSMax Reitz break; 1351d06107adSMax Reitz 1352d06107adSMax Reitz case MIRROR_METHOD_DISCARD: 13530b9fd3f4SFam Zheng ret = bdrv_co_pdiscard(bs->backing, offset, bytes); 1354d06107adSMax Reitz break; 1355d06107adSMax Reitz 1356d06107adSMax Reitz default: 1357d06107adSMax Reitz abort(); 1358d06107adSMax Reitz } 1359d06107adSMax Reitz 1360d06107adSMax Reitz if (ret < 0) { 1361d06107adSMax Reitz goto out; 1362d06107adSMax Reitz } 1363d06107adSMax Reitz 1364d06107adSMax Reitz if (copy_to_target) { 1365d06107adSMax Reitz do_sync_target_write(s->job, method, offset, bytes, qiov, flags); 1366d06107adSMax Reitz } 1367d06107adSMax Reitz 1368d06107adSMax Reitz out: 1369d06107adSMax Reitz if (copy_to_target) { 1370d06107adSMax Reitz active_write_settle(op); 1371d06107adSMax Reitz } 1372d06107adSMax Reitz return ret; 1373d06107adSMax Reitz } 1374d06107adSMax Reitz 13754ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, 13764ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 13774ef85a9cSKevin Wolf { 1378d06107adSMax Reitz MirrorBDSOpaque *s = bs->opaque; 1379d06107adSMax Reitz QEMUIOVector bounce_qiov; 1380d06107adSMax Reitz void *bounce_buf; 1381d06107adSMax Reitz int ret = 0; 1382d06107adSMax Reitz bool copy_to_target; 1383d06107adSMax Reitz 1384d06107adSMax Reitz copy_to_target = s->job->ret >= 0 && 1385d06107adSMax Reitz s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; 1386d06107adSMax Reitz 1387d06107adSMax Reitz if (copy_to_target) { 1388d06107adSMax Reitz /* The guest might concurrently modify the data to write; but 1389d06107adSMax Reitz * the data on source and destination must match, so we have 1390d06107adSMax Reitz * to use a bounce buffer if we are going to write to the 1391d06107adSMax Reitz * target now. */ 1392d06107adSMax Reitz bounce_buf = qemu_blockalign(bs, bytes); 1393d06107adSMax Reitz iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); 1394d06107adSMax Reitz 1395d06107adSMax Reitz qemu_iovec_init(&bounce_qiov, 1); 1396d06107adSMax Reitz qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); 1397d06107adSMax Reitz qiov = &bounce_qiov; 1398d06107adSMax Reitz } 1399d06107adSMax Reitz 1400d06107adSMax Reitz ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov, 1401d06107adSMax Reitz flags); 1402d06107adSMax Reitz 1403d06107adSMax Reitz if (copy_to_target) { 1404d06107adSMax Reitz qemu_iovec_destroy(&bounce_qiov); 1405d06107adSMax Reitz qemu_vfree(bounce_buf); 1406d06107adSMax Reitz } 1407d06107adSMax Reitz 1408d06107adSMax Reitz return ret; 14094ef85a9cSKevin Wolf } 14104ef85a9cSKevin Wolf 14114ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) 14124ef85a9cSKevin Wolf { 1413ce960aa9SVladimir Sementsov-Ogievskiy if (bs->backing == NULL) { 1414ce960aa9SVladimir Sementsov-Ogievskiy /* we can be here after failed bdrv_append in mirror_start_job */ 1415ce960aa9SVladimir Sementsov-Ogievskiy return 0; 1416ce960aa9SVladimir Sementsov-Ogievskiy } 14174ef85a9cSKevin Wolf return bdrv_co_flush(bs->backing->bs); 14184ef85a9cSKevin Wolf } 14194ef85a9cSKevin Wolf 14204ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, 1421f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 14224ef85a9cSKevin Wolf { 1423d06107adSMax Reitz return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL, 1424d06107adSMax Reitz flags); 14254ef85a9cSKevin Wolf } 14264ef85a9cSKevin Wolf 14274ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, 1428f5a5ca79SManos Pitsidianakis int64_t offset, int bytes) 14294ef85a9cSKevin Wolf { 1430d06107adSMax Reitz return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes, 1431d06107adSMax Reitz NULL, 0); 14324ef85a9cSKevin Wolf } 14334ef85a9cSKevin Wolf 1434fd4a6493SKevin Wolf static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) 1435fd4a6493SKevin Wolf { 143618775ff3SVladimir Sementsov-Ogievskiy if (bs->backing == NULL) { 143718775ff3SVladimir Sementsov-Ogievskiy /* we can be here after failed bdrv_attach_child in 143818775ff3SVladimir Sementsov-Ogievskiy * bdrv_set_backing_hd */ 143918775ff3SVladimir Sementsov-Ogievskiy return; 144018775ff3SVladimir Sementsov-Ogievskiy } 1441fd4a6493SKevin Wolf bdrv_refresh_filename(bs->backing->bs); 1442fd4a6493SKevin Wolf pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), 1443fd4a6493SKevin Wolf bs->backing->bs->filename); 1444fd4a6493SKevin Wolf } 1445fd4a6493SKevin Wolf 14464ef85a9cSKevin Wolf static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, 14474ef85a9cSKevin Wolf const BdrvChildRole *role, 1448e0995dc3SKevin Wolf BlockReopenQueue *reopen_queue, 14494ef85a9cSKevin Wolf uint64_t perm, uint64_t shared, 14504ef85a9cSKevin Wolf uint64_t *nperm, uint64_t *nshared) 14514ef85a9cSKevin Wolf { 14524ef85a9cSKevin Wolf /* Must be able to forward guest writes to the real image */ 14534ef85a9cSKevin Wolf *nperm = 0; 14544ef85a9cSKevin Wolf if (perm & BLK_PERM_WRITE) { 14554ef85a9cSKevin Wolf *nperm |= BLK_PERM_WRITE; 14564ef85a9cSKevin Wolf } 14574ef85a9cSKevin Wolf 14584ef85a9cSKevin Wolf *nshared = BLK_PERM_ALL; 14594ef85a9cSKevin Wolf } 14604ef85a9cSKevin Wolf 14614ef85a9cSKevin Wolf /* Dummy node that provides consistent read to its users without requiring it 14624ef85a9cSKevin Wolf * from its backing file and that allows writes on the backing file chain. */ 14634ef85a9cSKevin Wolf static BlockDriver bdrv_mirror_top = { 14644ef85a9cSKevin Wolf .format_name = "mirror_top", 14654ef85a9cSKevin Wolf .bdrv_co_preadv = bdrv_mirror_top_preadv, 14664ef85a9cSKevin Wolf .bdrv_co_pwritev = bdrv_mirror_top_pwritev, 14674ef85a9cSKevin Wolf .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, 14684ef85a9cSKevin Wolf .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, 14694ef85a9cSKevin Wolf .bdrv_co_flush = bdrv_mirror_top_flush, 14703e4d0e72SEric Blake .bdrv_co_block_status = bdrv_co_block_status_from_backing, 1471fd4a6493SKevin Wolf .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, 14724ef85a9cSKevin Wolf .bdrv_child_perm = bdrv_mirror_top_child_perm, 14734ef85a9cSKevin Wolf }; 14744ef85a9cSKevin Wolf 147571aa9867SAlberto Garcia static void mirror_start_job(const char *job_id, BlockDriverState *bs, 147647970dfbSJohn Snow int creation_flags, BlockDriverState *target, 147747970dfbSJohn Snow const char *replaces, int64_t speed, 147847970dfbSJohn Snow uint32_t granularity, int64_t buf_size, 1479274fcceeSMax Reitz BlockMirrorBackingMode backing_mode, 148003544a6eSFam Zheng BlockdevOnError on_source_error, 1481b952b558SPaolo Bonzini BlockdevOnError on_target_error, 14820fc9f8eaSFam Zheng bool unmap, 1483097310b5SMarkus Armbruster BlockCompletionFunc *cb, 148451ccfa2dSFam Zheng void *opaque, 148503544a6eSFam Zheng const BlockJobDriver *driver, 1486b49f7eadSWen Congyang bool is_none_mode, BlockDriverState *base, 148751ccfa2dSFam Zheng bool auto_complete, const char *filter_node_name, 1488481debaaSMax Reitz bool is_mirror, MirrorCopyMode copy_mode, 148951ccfa2dSFam Zheng Error **errp) 1490893f7ebaSPaolo Bonzini { 1491893f7ebaSPaolo Bonzini MirrorBlockJob *s; 1492429076e8SMax Reitz MirrorBDSOpaque *bs_opaque; 14934ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 14944ef85a9cSKevin Wolf bool target_graph_mod; 14954ef85a9cSKevin Wolf bool target_is_backing; 1496b2c2832cSKevin Wolf Error *local_err = NULL; 1497d7086422SKevin Wolf int ret; 1498893f7ebaSPaolo Bonzini 1499eee13dfeSPaolo Bonzini if (granularity == 0) { 1500341ebc2fSJohn Snow granularity = bdrv_get_default_bitmap_granularity(target); 1501eee13dfeSPaolo Bonzini } 1502eee13dfeSPaolo Bonzini 150331826642SEric Blake assert(is_power_of_2(granularity)); 1504eee13dfeSPaolo Bonzini 150548ac0a4dSWen Congyang if (buf_size < 0) { 150648ac0a4dSWen Congyang error_setg(errp, "Invalid parameter 'buf-size'"); 150748ac0a4dSWen Congyang return; 150848ac0a4dSWen Congyang } 150948ac0a4dSWen Congyang 151048ac0a4dSWen Congyang if (buf_size == 0) { 151148ac0a4dSWen Congyang buf_size = DEFAULT_MIRROR_BUF_SIZE; 151248ac0a4dSWen Congyang } 15135bc361b8SFam Zheng 151486fae10cSKevin Wolf if (bs == target) { 151586fae10cSKevin Wolf error_setg(errp, "Can't mirror node into itself"); 151686fae10cSKevin Wolf return; 151786fae10cSKevin Wolf } 151886fae10cSKevin Wolf 15194ef85a9cSKevin Wolf /* In the case of active commit, add dummy driver to provide consistent 15204ef85a9cSKevin Wolf * reads on the top, while disabling it in the intermediate nodes, and make 15214ef85a9cSKevin Wolf * the backing chain writable. */ 15226cdbceb1SKevin Wolf mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, 15236cdbceb1SKevin Wolf BDRV_O_RDWR, errp); 15244ef85a9cSKevin Wolf if (mirror_top_bs == NULL) { 1525893f7ebaSPaolo Bonzini return; 1526893f7ebaSPaolo Bonzini } 1527d3c8c674SKevin Wolf if (!filter_node_name) { 1528d3c8c674SKevin Wolf mirror_top_bs->implicit = true; 1529d3c8c674SKevin Wolf } 15304ef85a9cSKevin Wolf mirror_top_bs->total_sectors = bs->total_sectors; 1531228345bfSMax Reitz mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; 1532228345bfSMax Reitz mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED; 1533429076e8SMax Reitz bs_opaque = g_new0(MirrorBDSOpaque, 1); 1534429076e8SMax Reitz mirror_top_bs->opaque = bs_opaque; 153519dd29e8SFam Zheng bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); 1536893f7ebaSPaolo Bonzini 15374ef85a9cSKevin Wolf /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep 15387a25fcd0SMax Reitz * it alive until block_job_create() succeeds even if bs has no parent. */ 15394ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 15404ef85a9cSKevin Wolf bdrv_drained_begin(bs); 1541b2c2832cSKevin Wolf bdrv_append(mirror_top_bs, bs, &local_err); 15424ef85a9cSKevin Wolf bdrv_drained_end(bs); 15434ef85a9cSKevin Wolf 1544b2c2832cSKevin Wolf if (local_err) { 1545b2c2832cSKevin Wolf bdrv_unref(mirror_top_bs); 1546b2c2832cSKevin Wolf error_propagate(errp, local_err); 1547b2c2832cSKevin Wolf return; 1548b2c2832cSKevin Wolf } 1549b2c2832cSKevin Wolf 15504ef85a9cSKevin Wolf /* Make sure that the source is not resized while the job is running */ 155175859b94SJohn Snow s = block_job_create(job_id, driver, NULL, mirror_top_bs, 15524ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ, 15534ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 15544ef85a9cSKevin Wolf BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, 15554ef85a9cSKevin Wolf creation_flags, cb, opaque, errp); 15564ef85a9cSKevin Wolf if (!s) { 15574ef85a9cSKevin Wolf goto fail; 15584ef85a9cSKevin Wolf } 1559429076e8SMax Reitz bs_opaque->job = s; 1560429076e8SMax Reitz 15617a25fcd0SMax Reitz /* The block job now has a reference to this node */ 15627a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 15637a25fcd0SMax Reitz 15644ef85a9cSKevin Wolf s->mirror_top_bs = mirror_top_bs; 15654ef85a9cSKevin Wolf 15664ef85a9cSKevin Wolf /* No resize for the target either; while the mirror is still running, a 15674ef85a9cSKevin Wolf * consistent read isn't necessarily possible. We could possibly allow 15684ef85a9cSKevin Wolf * writes and graph modifications, though it would likely defeat the 15694ef85a9cSKevin Wolf * purpose of a mirror, so leave them blocked for now. 15704ef85a9cSKevin Wolf * 15714ef85a9cSKevin Wolf * In the case of active commit, things look a bit different, though, 15724ef85a9cSKevin Wolf * because the target is an already populated backing file in active use. 15734ef85a9cSKevin Wolf * We can allow anything except resize there.*/ 15744ef85a9cSKevin Wolf target_is_backing = bdrv_chain_contains(bs, target); 15754ef85a9cSKevin Wolf target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); 15764ef85a9cSKevin Wolf s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | 15774ef85a9cSKevin Wolf (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), 15784ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | 15794ef85a9cSKevin Wolf (target_is_backing ? BLK_PERM_CONSISTENT_READ | 15804ef85a9cSKevin Wolf BLK_PERM_WRITE | 15814ef85a9cSKevin Wolf BLK_PERM_GRAPH_MOD : 0)); 1582d7086422SKevin Wolf ret = blk_insert_bs(s->target, target, errp); 1583d7086422SKevin Wolf if (ret < 0) { 15844ef85a9cSKevin Wolf goto fail; 1585d7086422SKevin Wolf } 1586045a2f82SFam Zheng if (is_mirror) { 1587045a2f82SFam Zheng /* XXX: Mirror target could be a NBD server of target QEMU in the case 1588045a2f82SFam Zheng * of non-shared block migration. To allow migration completion, we 1589045a2f82SFam Zheng * have to allow "inactivate" of the target BB. When that happens, we 1590045a2f82SFam Zheng * know the job is drained, and the vcpus are stopped, so no write 1591045a2f82SFam Zheng * operation will be performed. Block layer already has assertions to 1592045a2f82SFam Zheng * ensure that. */ 1593045a2f82SFam Zheng blk_set_force_allow_inactivate(s->target); 1594045a2f82SFam Zheng } 1595e253f4b8SKevin Wolf 159609158f00SBenoît Canet s->replaces = g_strdup(replaces); 1597b952b558SPaolo Bonzini s->on_source_error = on_source_error; 1598b952b558SPaolo Bonzini s->on_target_error = on_target_error; 159903544a6eSFam Zheng s->is_none_mode = is_none_mode; 1600274fcceeSMax Reitz s->backing_mode = backing_mode; 1601481debaaSMax Reitz s->copy_mode = copy_mode; 16025bc361b8SFam Zheng s->base = base; 1603eee13dfeSPaolo Bonzini s->granularity = granularity; 160448ac0a4dSWen Congyang s->buf_size = ROUND_UP(buf_size, granularity); 16050fc9f8eaSFam Zheng s->unmap = unmap; 1606b49f7eadSWen Congyang if (auto_complete) { 1607b49f7eadSWen Congyang s->should_complete = true; 1608b49f7eadSWen Congyang } 1609b812f671SPaolo Bonzini 16100db6e54aSFam Zheng s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); 1611b8afb520SFam Zheng if (!s->dirty_bitmap) { 161288f9d1b3SKevin Wolf goto fail; 1613b8afb520SFam Zheng } 161410f3cd15SAlberto Garcia 16154ef85a9cSKevin Wolf /* Required permissions are already taken with blk_new() */ 161676d554e2SKevin Wolf block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, 161776d554e2SKevin Wolf &error_abort); 161876d554e2SKevin Wolf 1619f3ede4b0SAlberto Garcia /* In commit_active_start() all intermediate nodes disappear, so 1620f3ede4b0SAlberto Garcia * any jobs in them must be blocked */ 16214ef85a9cSKevin Wolf if (target_is_backing) { 1622f3ede4b0SAlberto Garcia BlockDriverState *iter; 1623f3ede4b0SAlberto Garcia for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { 16244ef85a9cSKevin Wolf /* XXX BLK_PERM_WRITE needs to be allowed so we don't block 16254ef85a9cSKevin Wolf * ourselves at s->base (if writes are blocked for a node, they are 16264ef85a9cSKevin Wolf * also blocked for its backing file). The other options would be a 16274ef85a9cSKevin Wolf * second filter driver above s->base (== target). */ 16284ef85a9cSKevin Wolf ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, 16294ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, 16304ef85a9cSKevin Wolf errp); 16314ef85a9cSKevin Wolf if (ret < 0) { 16324ef85a9cSKevin Wolf goto fail; 16334ef85a9cSKevin Wolf } 1634f3ede4b0SAlberto Garcia } 1635f3ede4b0SAlberto Garcia } 163610f3cd15SAlberto Garcia 163712aa4082SMax Reitz QTAILQ_INIT(&s->ops_in_flight); 163812aa4082SMax Reitz 16395ccac6f1SJohn Snow trace_mirror_start(bs, s, opaque); 1640da01ff7fSKevin Wolf job_start(&s->common.job); 16414ef85a9cSKevin Wolf return; 16424ef85a9cSKevin Wolf 16434ef85a9cSKevin Wolf fail: 16444ef85a9cSKevin Wolf if (s) { 16457a25fcd0SMax Reitz /* Make sure this BDS does not go away until we have completed the graph 16467a25fcd0SMax Reitz * changes below */ 16477a25fcd0SMax Reitz bdrv_ref(mirror_top_bs); 16487a25fcd0SMax Reitz 16494ef85a9cSKevin Wolf g_free(s->replaces); 16504ef85a9cSKevin Wolf blk_unref(s->target); 1651429076e8SMax Reitz bs_opaque->job = NULL; 16524ad35181SKevin Wolf job_early_fail(&s->common.job); 16534ef85a9cSKevin Wolf } 16544ef85a9cSKevin Wolf 1655c1cef672SFam Zheng bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 1656c1cef672SFam Zheng &error_abort); 16575fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 16587a25fcd0SMax Reitz 16597a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 1660893f7ebaSPaolo Bonzini } 166103544a6eSFam Zheng 166271aa9867SAlberto Garcia void mirror_start(const char *job_id, BlockDriverState *bs, 166371aa9867SAlberto Garcia BlockDriverState *target, const char *replaces, 1664a1999b33SJohn Snow int creation_flags, int64_t speed, 1665a1999b33SJohn Snow uint32_t granularity, int64_t buf_size, 1666274fcceeSMax Reitz MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, 1667274fcceeSMax Reitz BlockdevOnError on_source_error, 166803544a6eSFam Zheng BlockdevOnError on_target_error, 1669481debaaSMax Reitz bool unmap, const char *filter_node_name, 1670481debaaSMax Reitz MirrorCopyMode copy_mode, Error **errp) 167103544a6eSFam Zheng { 167203544a6eSFam Zheng bool is_none_mode; 167303544a6eSFam Zheng BlockDriverState *base; 167403544a6eSFam Zheng 16754b80ab2bSJohn Snow if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { 16764b80ab2bSJohn Snow error_setg(errp, "Sync mode 'incremental' not supported"); 1677d58d8453SJohn Snow return; 1678d58d8453SJohn Snow } 167903544a6eSFam Zheng is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 1680760e0063SKevin Wolf base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; 1681a1999b33SJohn Snow mirror_start_job(job_id, bs, creation_flags, target, replaces, 1682274fcceeSMax Reitz speed, granularity, buf_size, backing_mode, 168351ccfa2dSFam Zheng on_source_error, on_target_error, unmap, NULL, NULL, 16846cdbceb1SKevin Wolf &mirror_job_driver, is_none_mode, base, false, 1685481debaaSMax Reitz filter_node_name, true, copy_mode, errp); 168603544a6eSFam Zheng } 168703544a6eSFam Zheng 1688fd62c609SAlberto Garcia void commit_active_start(const char *job_id, BlockDriverState *bs, 168947970dfbSJohn Snow BlockDriverState *base, int creation_flags, 169047970dfbSJohn Snow int64_t speed, BlockdevOnError on_error, 16910db832f4SKevin Wolf const char *filter_node_name, 169278bbd910SFam Zheng BlockCompletionFunc *cb, void *opaque, 169378bbd910SFam Zheng bool auto_complete, Error **errp) 169403544a6eSFam Zheng { 16951ba79388SAlberto Garcia bool base_read_only; 1696cc67f4d1SJeff Cody Error *local_err = NULL; 16974da83585SJeff Cody 16981ba79388SAlberto Garcia base_read_only = bdrv_is_read_only(base); 16994da83585SJeff Cody 17001ba79388SAlberto Garcia if (base_read_only) { 17011ba79388SAlberto Garcia if (bdrv_reopen_set_read_only(base, false, errp) < 0) { 170220a63d2cSFam Zheng return; 170320a63d2cSFam Zheng } 17041ba79388SAlberto Garcia } 17054da83585SJeff Cody 170647970dfbSJohn Snow mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, 170771aa9867SAlberto Garcia MIRROR_LEAVE_BACKING_CHAIN, 170851ccfa2dSFam Zheng on_error, on_error, true, cb, opaque, 17096cdbceb1SKevin Wolf &commit_active_job_driver, false, base, auto_complete, 1710481debaaSMax Reitz filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, 1711481debaaSMax Reitz &local_err); 17120fb6395cSMarkus Armbruster if (local_err) { 1713cc67f4d1SJeff Cody error_propagate(errp, local_err); 17144da83585SJeff Cody goto error_restore_flags; 17154da83585SJeff Cody } 17164da83585SJeff Cody 17174da83585SJeff Cody return; 17184da83585SJeff Cody 17194da83585SJeff Cody error_restore_flags: 17204da83585SJeff Cody /* ignore error and errp for bdrv_reopen, because we want to propagate 17214da83585SJeff Cody * the original error */ 17221ba79388SAlberto Garcia if (base_read_only) { 17231ba79388SAlberto Garcia bdrv_reopen_set_read_only(base, true, NULL); 17241ba79388SAlberto Garcia } 17254da83585SJeff Cody return; 172603544a6eSFam Zheng } 1727