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; 835e771752SSergio Lopez bool in_drain; 84893f7ebaSPaolo Bonzini } MirrorBlockJob; 85893f7ebaSPaolo Bonzini 86429076e8SMax Reitz typedef struct MirrorBDSOpaque { 87429076e8SMax Reitz MirrorBlockJob *job; 88f94dc3b4SMax Reitz bool stop; 89429076e8SMax Reitz } MirrorBDSOpaque; 90429076e8SMax Reitz 9112aa4082SMax Reitz struct MirrorOp { 92bd48bde8SPaolo Bonzini MirrorBlockJob *s; 93bd48bde8SPaolo Bonzini QEMUIOVector qiov; 94b436982fSEric Blake int64_t offset; 95b436982fSEric Blake uint64_t bytes; 962e1990b2SMax Reitz 972e1990b2SMax Reitz /* The pointee is set by mirror_co_read(), mirror_co_zero(), and 982e1990b2SMax Reitz * mirror_co_discard() before yielding for the first time */ 992e1990b2SMax Reitz int64_t *bytes_handled; 10012aa4082SMax Reitz 1011181e19aSMax Reitz bool is_pseudo_op; 102d06107adSMax Reitz bool is_active_write; 10312aa4082SMax Reitz CoQueue waiting_requests; 10412aa4082SMax Reitz 10512aa4082SMax Reitz QTAILQ_ENTRY(MirrorOp) next; 10612aa4082SMax Reitz }; 107bd48bde8SPaolo Bonzini 1084295c5fcSMax Reitz typedef enum MirrorMethod { 1094295c5fcSMax Reitz MIRROR_METHOD_COPY, 1104295c5fcSMax Reitz MIRROR_METHOD_ZERO, 1114295c5fcSMax Reitz MIRROR_METHOD_DISCARD, 1124295c5fcSMax Reitz } MirrorMethod; 1134295c5fcSMax Reitz 114b952b558SPaolo Bonzini static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 115b952b558SPaolo Bonzini int error) 116b952b558SPaolo Bonzini { 117b952b558SPaolo Bonzini s->synced = false; 118d06107adSMax Reitz s->actively_synced = false; 119b952b558SPaolo Bonzini if (read) { 12081e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_source_error, 12181e254dcSKevin Wolf true, error); 122b952b558SPaolo Bonzini } else { 12381e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_target_error, 12481e254dcSKevin Wolf false, error); 125b952b558SPaolo Bonzini } 126b952b558SPaolo Bonzini } 127b952b558SPaolo Bonzini 1281181e19aSMax Reitz static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, 1291181e19aSMax Reitz MirrorBlockJob *s, 1301181e19aSMax Reitz uint64_t offset, 1311181e19aSMax Reitz uint64_t bytes) 1321181e19aSMax Reitz { 1331181e19aSMax Reitz uint64_t self_start_chunk = offset / s->granularity; 1341181e19aSMax Reitz uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); 1351181e19aSMax Reitz uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; 1361181e19aSMax Reitz 1371181e19aSMax Reitz while (find_next_bit(s->in_flight_bitmap, self_end_chunk, 1381181e19aSMax Reitz self_start_chunk) < self_end_chunk && 1391181e19aSMax Reitz s->ret >= 0) 1401181e19aSMax Reitz { 1411181e19aSMax Reitz MirrorOp *op; 1421181e19aSMax Reitz 1431181e19aSMax Reitz QTAILQ_FOREACH(op, &s->ops_in_flight, next) { 1441181e19aSMax Reitz uint64_t op_start_chunk = op->offset / s->granularity; 1451181e19aSMax Reitz uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, 1461181e19aSMax Reitz s->granularity) - 1471181e19aSMax Reitz op_start_chunk; 1481181e19aSMax Reitz 1491181e19aSMax Reitz if (op == self) { 1501181e19aSMax Reitz continue; 1511181e19aSMax Reitz } 1521181e19aSMax Reitz 1531181e19aSMax Reitz if (ranges_overlap(self_start_chunk, self_nb_chunks, 1541181e19aSMax Reitz op_start_chunk, op_nb_chunks)) 1551181e19aSMax Reitz { 1561181e19aSMax Reitz qemu_co_queue_wait(&op->waiting_requests, NULL); 1571181e19aSMax Reitz break; 1581181e19aSMax Reitz } 1591181e19aSMax Reitz } 1601181e19aSMax Reitz } 1611181e19aSMax Reitz } 1621181e19aSMax Reitz 1632e1990b2SMax Reitz static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) 164bd48bde8SPaolo Bonzini { 165bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 166402a4741SPaolo Bonzini struct iovec *iov; 167bd48bde8SPaolo Bonzini int64_t chunk_num; 168b436982fSEric Blake int i, nb_chunks; 169bd48bde8SPaolo Bonzini 170b436982fSEric Blake trace_mirror_iteration_done(s, op->offset, op->bytes, ret); 171bd48bde8SPaolo Bonzini 172bd48bde8SPaolo Bonzini s->in_flight--; 173b436982fSEric Blake s->bytes_in_flight -= op->bytes; 174402a4741SPaolo Bonzini iov = op->qiov.iov; 175402a4741SPaolo Bonzini for (i = 0; i < op->qiov.niov; i++) { 176402a4741SPaolo Bonzini MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 177402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 178402a4741SPaolo Bonzini s->buf_free_count++; 179402a4741SPaolo Bonzini } 180402a4741SPaolo Bonzini 181b436982fSEric Blake chunk_num = op->offset / s->granularity; 182b436982fSEric Blake nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 18312aa4082SMax Reitz 184402a4741SPaolo Bonzini bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 18512aa4082SMax Reitz QTAILQ_REMOVE(&s->ops_in_flight, op, next); 186b21c7652SMax Reitz if (ret >= 0) { 187b21c7652SMax Reitz if (s->cow_bitmap) { 188bd48bde8SPaolo Bonzini bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 189bd48bde8SPaolo Bonzini } 19090ab48ebSAnton Nefedov if (!s->initial_zeroing_ongoing) { 19130a5c887SKevin Wolf job_progress_update(&s->common.job, op->bytes); 192b21c7652SMax Reitz } 19390ab48ebSAnton Nefedov } 1946df3bf8eSZhang Min qemu_iovec_destroy(&op->qiov); 1957b770c72SStefan Hajnoczi 19612aa4082SMax Reitz qemu_co_queue_restart_all(&op->waiting_requests); 19712aa4082SMax Reitz g_free(op); 1987b770c72SStefan Hajnoczi } 199bd48bde8SPaolo Bonzini 2002e1990b2SMax Reitz static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) 201bd48bde8SPaolo Bonzini { 202bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 203b9e413ddSPaolo Bonzini 204bd48bde8SPaolo Bonzini if (ret < 0) { 205bd48bde8SPaolo Bonzini BlockErrorAction action; 206bd48bde8SPaolo Bonzini 207e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 208bd48bde8SPaolo Bonzini action = mirror_error_action(s, false, -ret); 209a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 210bd48bde8SPaolo Bonzini s->ret = ret; 211bd48bde8SPaolo Bonzini } 212bd48bde8SPaolo Bonzini } 213d12ade57SVladimir Sementsov-Ogievskiy 214bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 215bd48bde8SPaolo Bonzini } 216bd48bde8SPaolo Bonzini 2172e1990b2SMax Reitz static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) 218bd48bde8SPaolo Bonzini { 219bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 220b9e413ddSPaolo Bonzini 221bd48bde8SPaolo Bonzini if (ret < 0) { 222bd48bde8SPaolo Bonzini BlockErrorAction action; 223bd48bde8SPaolo Bonzini 224e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 225bd48bde8SPaolo Bonzini action = mirror_error_action(s, true, -ret); 226a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 227bd48bde8SPaolo Bonzini s->ret = ret; 228bd48bde8SPaolo Bonzini } 229bd48bde8SPaolo Bonzini 230bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 231d12ade57SVladimir Sementsov-Ogievskiy return; 232bd48bde8SPaolo Bonzini } 233d12ade57SVladimir Sementsov-Ogievskiy 234d12ade57SVladimir Sementsov-Ogievskiy ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0); 235d12ade57SVladimir Sementsov-Ogievskiy mirror_write_complete(op, ret); 236b9e413ddSPaolo Bonzini } 237bd48bde8SPaolo Bonzini 238782d97efSEric Blake /* Clip bytes relative to offset to not exceed end-of-file */ 239782d97efSEric Blake static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, 240782d97efSEric Blake int64_t offset, 241782d97efSEric Blake int64_t bytes) 242782d97efSEric Blake { 243782d97efSEric Blake return MIN(bytes, s->bdev_length - offset); 244782d97efSEric Blake } 245782d97efSEric Blake 246782d97efSEric Blake /* Round offset and/or bytes to target cluster if COW is needed, and 247782d97efSEric Blake * return the offset of the adjusted tail against original. */ 248782d97efSEric Blake static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, 249ae4cc877SEric Blake uint64_t *bytes) 250893f7ebaSPaolo Bonzini { 251e5b43573SFam Zheng bool need_cow; 252e5b43573SFam Zheng int ret = 0; 253782d97efSEric Blake int64_t align_offset = *offset; 2547cfd5275SEric Blake int64_t align_bytes = *bytes; 255782d97efSEric Blake int max_bytes = s->granularity * s->max_iov; 256893f7ebaSPaolo Bonzini 257782d97efSEric Blake need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); 258782d97efSEric Blake need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, 259e5b43573SFam Zheng s->cow_bitmap); 260e5b43573SFam Zheng if (need_cow) { 261782d97efSEric Blake bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, 262782d97efSEric Blake &align_offset, &align_bytes); 2638f0720ecSPaolo Bonzini } 2648f0720ecSPaolo Bonzini 265782d97efSEric Blake if (align_bytes > max_bytes) { 266782d97efSEric Blake align_bytes = max_bytes; 267e5b43573SFam Zheng if (need_cow) { 268782d97efSEric Blake align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); 269e5b43573SFam Zheng } 270e5b43573SFam Zheng } 271782d97efSEric Blake /* Clipping may result in align_bytes unaligned to chunk boundary, but 2724150ae60SFam Zheng * that doesn't matter because it's already the end of source image. */ 273782d97efSEric Blake align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); 274402a4741SPaolo Bonzini 275782d97efSEric Blake ret = align_offset + align_bytes - (*offset + *bytes); 276782d97efSEric Blake *offset = align_offset; 277782d97efSEric Blake *bytes = align_bytes; 278e5b43573SFam Zheng assert(ret >= 0); 279e5b43573SFam Zheng return ret; 280e5b43573SFam Zheng } 281e5b43573SFam Zheng 282537c3d4fSStefan Hajnoczi static inline void coroutine_fn 283537c3d4fSStefan Hajnoczi mirror_wait_for_any_operation(MirrorBlockJob *s, bool active) 28421cd917fSFam Zheng { 28512aa4082SMax Reitz MirrorOp *op; 28612aa4082SMax Reitz 2871181e19aSMax Reitz QTAILQ_FOREACH(op, &s->ops_in_flight, next) { 2881181e19aSMax Reitz /* Do not wait on pseudo ops, because it may in turn wait on 2891181e19aSMax Reitz * some other operation to start, which may in fact be the 2901181e19aSMax Reitz * caller of this function. Since there is only one pseudo op 2911181e19aSMax Reitz * at any given time, we will always find some real operation 2921181e19aSMax Reitz * to wait on. */ 293d06107adSMax Reitz if (!op->is_pseudo_op && op->is_active_write == active) { 29412aa4082SMax Reitz qemu_co_queue_wait(&op->waiting_requests, NULL); 2951181e19aSMax Reitz return; 2961181e19aSMax Reitz } 2971181e19aSMax Reitz } 2981181e19aSMax Reitz abort(); 29921cd917fSFam Zheng } 30021cd917fSFam Zheng 301537c3d4fSStefan Hajnoczi static inline void coroutine_fn 302537c3d4fSStefan Hajnoczi mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) 303d06107adSMax Reitz { 304d06107adSMax Reitz /* Only non-active operations use up in-flight slots */ 305d06107adSMax Reitz mirror_wait_for_any_operation(s, false); 306d06107adSMax Reitz } 307d06107adSMax Reitz 3082e1990b2SMax Reitz /* Perform a mirror copy operation. 3092e1990b2SMax Reitz * 3102e1990b2SMax Reitz * *op->bytes_handled is set to the number of bytes copied after and 3112e1990b2SMax Reitz * including offset, excluding any bytes copied prior to offset due 3122e1990b2SMax Reitz * to alignment. This will be op->bytes if no alignment is necessary, 3132e1990b2SMax Reitz * or (new_end - op->offset) if the tail is rounded up or down due to 314e5b43573SFam Zheng * alignment or buffer limit. 315402a4741SPaolo Bonzini */ 3162e1990b2SMax Reitz static void coroutine_fn mirror_co_read(void *opaque) 317e5b43573SFam Zheng { 3182e1990b2SMax Reitz MirrorOp *op = opaque; 3192e1990b2SMax Reitz MirrorBlockJob *s = op->s; 320ae4cc877SEric Blake int nb_chunks; 321ae4cc877SEric Blake uint64_t ret; 322ae4cc877SEric Blake uint64_t max_bytes; 323402a4741SPaolo Bonzini 324ae4cc877SEric Blake max_bytes = s->granularity * s->max_iov; 325e5b43573SFam Zheng 326e5b43573SFam Zheng /* We can only handle as much as buf_size at a time. */ 3272e1990b2SMax Reitz op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); 3282e1990b2SMax Reitz assert(op->bytes); 3292e1990b2SMax Reitz assert(op->bytes < BDRV_REQUEST_MAX_BYTES); 3302e1990b2SMax Reitz *op->bytes_handled = op->bytes; 331e5b43573SFam Zheng 332e5b43573SFam Zheng if (s->cow_bitmap) { 3332e1990b2SMax Reitz *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); 334e5b43573SFam Zheng } 3352e1990b2SMax Reitz /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ 3362e1990b2SMax Reitz assert(*op->bytes_handled <= UINT_MAX); 3372e1990b2SMax Reitz assert(op->bytes <= s->buf_size); 338ae4cc877SEric Blake /* The offset is granularity-aligned because: 339e5b43573SFam Zheng * 1) Caller passes in aligned values; 340e5b43573SFam Zheng * 2) mirror_cow_align is used only when target cluster is larger. */ 3412e1990b2SMax Reitz assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); 342ae4cc877SEric Blake /* The range is sector-aligned, since bdrv_getlength() rounds up. */ 3432e1990b2SMax Reitz assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); 3442e1990b2SMax Reitz nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 345e5b43573SFam Zheng 346e5b43573SFam Zheng while (s->buf_free_count < nb_chunks) { 3472e1990b2SMax Reitz trace_mirror_yield_in_flight(s, op->offset, s->in_flight); 3481181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 349b812f671SPaolo Bonzini } 350b812f671SPaolo Bonzini 351402a4741SPaolo Bonzini /* Now make a QEMUIOVector taking enough granularity-sized chunks 352402a4741SPaolo Bonzini * from s->buf_free. 353402a4741SPaolo Bonzini */ 354402a4741SPaolo Bonzini qemu_iovec_init(&op->qiov, nb_chunks); 355402a4741SPaolo Bonzini while (nb_chunks-- > 0) { 356402a4741SPaolo Bonzini MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 3572e1990b2SMax Reitz size_t remaining = op->bytes - op->qiov.size; 3585a0f6fd5SKevin Wolf 359402a4741SPaolo Bonzini QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 360402a4741SPaolo Bonzini s->buf_free_count--; 3615a0f6fd5SKevin Wolf qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 362402a4741SPaolo Bonzini } 363402a4741SPaolo Bonzini 364893f7ebaSPaolo Bonzini /* Copy the dirty cluster. */ 365bd48bde8SPaolo Bonzini s->in_flight++; 3662e1990b2SMax Reitz s->bytes_in_flight += op->bytes; 3672e1990b2SMax Reitz trace_mirror_one_iteration(s, op->offset, op->bytes); 368dcfb3bebSFam Zheng 369138f9fffSMax Reitz ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, 370138f9fffSMax Reitz &op->qiov, 0); 3712e1990b2SMax Reitz mirror_read_complete(op, ret); 372e5b43573SFam Zheng } 373e5b43573SFam Zheng 3742e1990b2SMax Reitz static void coroutine_fn mirror_co_zero(void *opaque) 375e5b43573SFam Zheng { 3762e1990b2SMax Reitz MirrorOp *op = opaque; 3772e1990b2SMax Reitz int ret; 378e5b43573SFam Zheng 3792e1990b2SMax Reitz op->s->in_flight++; 3802e1990b2SMax Reitz op->s->bytes_in_flight += op->bytes; 3812e1990b2SMax Reitz *op->bytes_handled = op->bytes; 382e5b43573SFam Zheng 3832e1990b2SMax Reitz ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, 3842e1990b2SMax Reitz op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); 3852e1990b2SMax Reitz mirror_write_complete(op, ret); 386e5b43573SFam Zheng } 3872e1990b2SMax Reitz 3882e1990b2SMax Reitz static void coroutine_fn mirror_co_discard(void *opaque) 3892e1990b2SMax Reitz { 3902e1990b2SMax Reitz MirrorOp *op = opaque; 3912e1990b2SMax Reitz int ret; 3922e1990b2SMax Reitz 3932e1990b2SMax Reitz op->s->in_flight++; 3942e1990b2SMax Reitz op->s->bytes_in_flight += op->bytes; 3952e1990b2SMax Reitz *op->bytes_handled = op->bytes; 3962e1990b2SMax Reitz 3972e1990b2SMax Reitz ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); 3982e1990b2SMax Reitz mirror_write_complete(op, ret); 399e5b43573SFam Zheng } 400e5b43573SFam Zheng 4014295c5fcSMax Reitz static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, 4024295c5fcSMax Reitz unsigned bytes, MirrorMethod mirror_method) 4034295c5fcSMax Reitz { 4042e1990b2SMax Reitz MirrorOp *op; 4052e1990b2SMax Reitz Coroutine *co; 4062e1990b2SMax Reitz int64_t bytes_handled = -1; 4072e1990b2SMax Reitz 4082e1990b2SMax Reitz op = g_new(MirrorOp, 1); 4092e1990b2SMax Reitz *op = (MirrorOp){ 4102e1990b2SMax Reitz .s = s, 4112e1990b2SMax Reitz .offset = offset, 4122e1990b2SMax Reitz .bytes = bytes, 4132e1990b2SMax Reitz .bytes_handled = &bytes_handled, 4142e1990b2SMax Reitz }; 41512aa4082SMax Reitz qemu_co_queue_init(&op->waiting_requests); 4162e1990b2SMax Reitz 4174295c5fcSMax Reitz switch (mirror_method) { 4184295c5fcSMax Reitz case MIRROR_METHOD_COPY: 4192e1990b2SMax Reitz co = qemu_coroutine_create(mirror_co_read, op); 4202e1990b2SMax Reitz break; 4214295c5fcSMax Reitz case MIRROR_METHOD_ZERO: 4222e1990b2SMax Reitz co = qemu_coroutine_create(mirror_co_zero, op); 4232e1990b2SMax Reitz break; 4244295c5fcSMax Reitz case MIRROR_METHOD_DISCARD: 4252e1990b2SMax Reitz co = qemu_coroutine_create(mirror_co_discard, op); 4262e1990b2SMax Reitz break; 4274295c5fcSMax Reitz default: 4284295c5fcSMax Reitz abort(); 4294295c5fcSMax Reitz } 4302e1990b2SMax Reitz 43112aa4082SMax Reitz QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); 4322e1990b2SMax Reitz qemu_coroutine_enter(co); 4332e1990b2SMax Reitz /* At this point, ownership of op has been moved to the coroutine 4342e1990b2SMax Reitz * and the object may already be freed */ 4352e1990b2SMax Reitz 4362e1990b2SMax Reitz /* Assert that this value has been set */ 4372e1990b2SMax Reitz assert(bytes_handled >= 0); 4382e1990b2SMax Reitz 4392e1990b2SMax Reitz /* Same assertion as in mirror_co_read() (and for mirror_co_read() 4402e1990b2SMax Reitz * and mirror_co_discard(), bytes_handled == op->bytes, which 4412e1990b2SMax Reitz * is the @bytes parameter given to this function) */ 4422e1990b2SMax Reitz assert(bytes_handled <= UINT_MAX); 4432e1990b2SMax Reitz return bytes_handled; 4444295c5fcSMax Reitz } 4454295c5fcSMax Reitz 446e5b43573SFam Zheng static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) 447e5b43573SFam Zheng { 448138f9fffSMax Reitz BlockDriverState *source = s->mirror_top_bs->backing->bs; 4491181e19aSMax Reitz MirrorOp *pseudo_op; 4501181e19aSMax Reitz int64_t offset; 4511181e19aSMax Reitz uint64_t delay_ns = 0, ret = 0; 452e5b43573SFam Zheng /* At least the first dirty chunk is mirrored in one iteration. */ 453e5b43573SFam Zheng int nb_chunks = 1; 4544b5004d9SDenis V. Lunev bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); 455b436982fSEric Blake int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); 456e5b43573SFam Zheng 457b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_lock(s->dirty_bitmap); 458f798184cSEric Blake offset = bdrv_dirty_iter_next(s->dbi); 459fb2ef791SEric Blake if (offset < 0) { 460dc162c8eSFam Zheng bdrv_set_dirty_iter(s->dbi, 0); 461f798184cSEric Blake offset = bdrv_dirty_iter_next(s->dbi); 4629a46dba7SEric Blake trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 463fb2ef791SEric Blake assert(offset >= 0); 464e5b43573SFam Zheng } 465b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 466e5b43573SFam Zheng 4671181e19aSMax Reitz mirror_wait_on_conflicts(NULL, s, offset, 1); 4689c83625bSMax Reitz 469da01ff7fSKevin Wolf job_pause_point(&s->common.job); 470565ac01fSStefan Hajnoczi 471e5b43573SFam Zheng /* Find the number of consective dirty chunks following the first dirty 472e5b43573SFam Zheng * one, and wait for in flight requests in them. */ 473b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_lock(s->dirty_bitmap); 474fb2ef791SEric Blake while (nb_chunks * s->granularity < s->buf_size) { 475dc162c8eSFam Zheng int64_t next_dirty; 476fb2ef791SEric Blake int64_t next_offset = offset + nb_chunks * s->granularity; 477fb2ef791SEric Blake int64_t next_chunk = next_offset / s->granularity; 478fb2ef791SEric Blake if (next_offset >= s->bdev_length || 4793b5d4df0SEric Blake !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { 480e5b43573SFam Zheng break; 481e5b43573SFam Zheng } 482e5b43573SFam Zheng if (test_bit(next_chunk, s->in_flight_bitmap)) { 483e5b43573SFam Zheng break; 484e5b43573SFam Zheng } 4859c83625bSMax Reitz 486f798184cSEric Blake next_dirty = bdrv_dirty_iter_next(s->dbi); 487fb2ef791SEric Blake if (next_dirty > next_offset || next_dirty < 0) { 488f27a2742SMax Reitz /* The bitmap iterator's cache is stale, refresh it */ 489715a74d8SEric Blake bdrv_set_dirty_iter(s->dbi, next_offset); 490f798184cSEric Blake next_dirty = bdrv_dirty_iter_next(s->dbi); 491f27a2742SMax Reitz } 492fb2ef791SEric Blake assert(next_dirty == next_offset); 493e5b43573SFam Zheng nb_chunks++; 494e5b43573SFam Zheng } 495e5b43573SFam Zheng 496e5b43573SFam Zheng /* Clear dirty bits before querying the block status, because 49731826642SEric Blake * calling bdrv_block_status_above could yield - if some blocks are 498e5b43573SFam Zheng * marked dirty in this window, we need to know. 499e5b43573SFam Zheng */ 500e0d7f73eSEric Blake bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, 501e0d7f73eSEric Blake nb_chunks * s->granularity); 502b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 503b64bd51eSPaolo Bonzini 5041181e19aSMax Reitz /* Before claiming an area in the in-flight bitmap, we have to 5051181e19aSMax Reitz * create a MirrorOp for it so that conflicting requests can wait 5061181e19aSMax Reitz * for it. mirror_perform() will create the real MirrorOps later, 5071181e19aSMax Reitz * for now we just create a pseudo operation that will wake up all 5081181e19aSMax Reitz * conflicting requests once all real operations have been 5091181e19aSMax Reitz * launched. */ 5101181e19aSMax Reitz pseudo_op = g_new(MirrorOp, 1); 5111181e19aSMax Reitz *pseudo_op = (MirrorOp){ 5121181e19aSMax Reitz .offset = offset, 5131181e19aSMax Reitz .bytes = nb_chunks * s->granularity, 5141181e19aSMax Reitz .is_pseudo_op = true, 5151181e19aSMax Reitz }; 5161181e19aSMax Reitz qemu_co_queue_init(&pseudo_op->waiting_requests); 5171181e19aSMax Reitz QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); 5181181e19aSMax Reitz 519fb2ef791SEric Blake bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); 520fb2ef791SEric Blake while (nb_chunks > 0 && offset < s->bdev_length) { 52131826642SEric Blake int ret; 5227cfd5275SEric Blake int64_t io_bytes; 523f3e4ce4aSEric Blake int64_t io_bytes_acct; 5244295c5fcSMax Reitz MirrorMethod mirror_method = MIRROR_METHOD_COPY; 525e5b43573SFam Zheng 526fb2ef791SEric Blake assert(!(offset % s->granularity)); 52731826642SEric Blake ret = bdrv_block_status_above(source, NULL, offset, 52831826642SEric Blake nb_chunks * s->granularity, 52931826642SEric Blake &io_bytes, NULL, NULL); 530e5b43573SFam Zheng if (ret < 0) { 531fb2ef791SEric Blake io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); 5320965a41eSVladimir Sementsov-Ogievskiy } else if (ret & BDRV_BLOCK_DATA) { 533fb2ef791SEric Blake io_bytes = MIN(io_bytes, max_io_bytes); 534e5b43573SFam Zheng } 535e5b43573SFam Zheng 536fb2ef791SEric Blake io_bytes -= io_bytes % s->granularity; 537fb2ef791SEric Blake if (io_bytes < s->granularity) { 538fb2ef791SEric Blake io_bytes = s->granularity; 539e5b43573SFam Zheng } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { 540fb2ef791SEric Blake int64_t target_offset; 5417cfd5275SEric Blake int64_t target_bytes; 542fb2ef791SEric Blake bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, 543fb2ef791SEric Blake &target_offset, &target_bytes); 544fb2ef791SEric Blake if (target_offset == offset && 545fb2ef791SEric Blake target_bytes == io_bytes) { 546e5b43573SFam Zheng mirror_method = ret & BDRV_BLOCK_ZERO ? 547e5b43573SFam Zheng MIRROR_METHOD_ZERO : 548e5b43573SFam Zheng MIRROR_METHOD_DISCARD; 549e5b43573SFam Zheng } 550e5b43573SFam Zheng } 551e5b43573SFam Zheng 552cf56a3c6SDenis V. Lunev while (s->in_flight >= MAX_IN_FLIGHT) { 553fb2ef791SEric Blake trace_mirror_yield_in_flight(s, offset, s->in_flight); 5541181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 555cf56a3c6SDenis V. Lunev } 556cf56a3c6SDenis V. Lunev 557dbaa7b57SVladimir Sementsov-Ogievskiy if (s->ret < 0) { 5581181e19aSMax Reitz ret = 0; 5591181e19aSMax Reitz goto fail; 560dbaa7b57SVladimir Sementsov-Ogievskiy } 561dbaa7b57SVladimir Sementsov-Ogievskiy 562fb2ef791SEric Blake io_bytes = mirror_clip_bytes(s, offset, io_bytes); 5634295c5fcSMax Reitz io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); 5644295c5fcSMax Reitz if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { 565f3e4ce4aSEric Blake io_bytes_acct = 0; 5664b5004d9SDenis V. Lunev } else { 567fb2ef791SEric Blake io_bytes_acct = io_bytes; 5684b5004d9SDenis V. Lunev } 569fb2ef791SEric Blake assert(io_bytes); 570fb2ef791SEric Blake offset += io_bytes; 571fb2ef791SEric Blake nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); 572dee81d51SKevin Wolf delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct); 573dcfb3bebSFam Zheng } 5741181e19aSMax Reitz 5751181e19aSMax Reitz ret = delay_ns; 5761181e19aSMax Reitz fail: 5771181e19aSMax Reitz QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); 5781181e19aSMax Reitz qemu_co_queue_restart_all(&pseudo_op->waiting_requests); 5791181e19aSMax Reitz g_free(pseudo_op); 5801181e19aSMax Reitz 5811181e19aSMax Reitz return ret; 582893f7ebaSPaolo Bonzini } 583b952b558SPaolo Bonzini 584402a4741SPaolo Bonzini static void mirror_free_init(MirrorBlockJob *s) 585402a4741SPaolo Bonzini { 586402a4741SPaolo Bonzini int granularity = s->granularity; 587402a4741SPaolo Bonzini size_t buf_size = s->buf_size; 588402a4741SPaolo Bonzini uint8_t *buf = s->buf; 589402a4741SPaolo Bonzini 590402a4741SPaolo Bonzini assert(s->buf_free_count == 0); 591402a4741SPaolo Bonzini QSIMPLEQ_INIT(&s->buf_free); 592402a4741SPaolo Bonzini while (buf_size != 0) { 593402a4741SPaolo Bonzini MirrorBuffer *cur = (MirrorBuffer *)buf; 594402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 595402a4741SPaolo Bonzini s->buf_free_count++; 596402a4741SPaolo Bonzini buf_size -= granularity; 597402a4741SPaolo Bonzini buf += granularity; 598402a4741SPaolo Bonzini } 599402a4741SPaolo Bonzini } 600402a4741SPaolo Bonzini 601bae8196dSPaolo Bonzini /* This is also used for the .pause callback. There is no matching 602bae8196dSPaolo Bonzini * mirror_resume() because mirror_run() will begin iterating again 603bae8196dSPaolo Bonzini * when the job is resumed. 604bae8196dSPaolo Bonzini */ 605537c3d4fSStefan Hajnoczi static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s) 606bd48bde8SPaolo Bonzini { 607bd48bde8SPaolo Bonzini while (s->in_flight > 0) { 6081181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 609bd48bde8SPaolo Bonzini } 610893f7ebaSPaolo Bonzini } 611893f7ebaSPaolo Bonzini 612737efc1eSJohn Snow /** 613737efc1eSJohn Snow * mirror_exit_common: handle both abort() and prepare() cases. 614737efc1eSJohn Snow * for .prepare, returns 0 on success and -errno on failure. 615737efc1eSJohn Snow * for .abort cases, denoted by abort = true, MUST return 0. 616737efc1eSJohn Snow */ 617737efc1eSJohn Snow static int mirror_exit_common(Job *job) 6185a7e7a0bSStefan Hajnoczi { 6191908a559SKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 6201908a559SKevin Wolf BlockJob *bjob = &s->common; 621429076e8SMax Reitz MirrorBDSOpaque *bs_opaque = s->mirror_top_bs->opaque; 6225a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context = NULL; 623138f9fffSMax Reitz BlockDriverState *src = s->mirror_top_bs->backing->bs; 624e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 6254ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs = s->mirror_top_bs; 62612fa4af6SKevin Wolf Error *local_err = NULL; 627737efc1eSJohn Snow bool abort = job->ret < 0; 628737efc1eSJohn Snow int ret = 0; 629737efc1eSJohn Snow 630737efc1eSJohn Snow if (s->prepared) { 631737efc1eSJohn Snow return 0; 632737efc1eSJohn Snow } 633737efc1eSJohn Snow s->prepared = true; 6343f09bfbcSKevin Wolf 635ef53dc09SAlberto Garcia if (bdrv_chain_contains(src, target_bs)) { 636ef53dc09SAlberto Garcia bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs); 637ef53dc09SAlberto Garcia } 638ef53dc09SAlberto Garcia 6392119882cSPaolo Bonzini bdrv_release_dirty_bitmap(src, s->dirty_bitmap); 6402119882cSPaolo Bonzini 6417b508f6bSJohn Snow /* Make sure that the source BDS doesn't go away during bdrv_replace_node, 6427b508f6bSJohn Snow * before we can call bdrv_drained_end */ 6433f09bfbcSKevin Wolf bdrv_ref(src); 6444ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 6457d9fcb39SKevin Wolf bdrv_ref(target_bs); 6467d9fcb39SKevin Wolf 6477d9fcb39SKevin Wolf /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before 6487d9fcb39SKevin Wolf * inserting target_bs at s->to_replace, where we might not be able to get 64963c8ef28SKevin Wolf * these permissions. 65063c8ef28SKevin Wolf * 65163c8ef28SKevin Wolf * Note that blk_unref() alone doesn't necessarily drop permissions because 65263c8ef28SKevin Wolf * we might be running nested inside mirror_drain(), which takes an extra 65363c8ef28SKevin Wolf * reference, so use an explicit blk_set_perm() first. */ 65463c8ef28SKevin Wolf blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); 6557d9fcb39SKevin Wolf blk_unref(s->target); 6567d9fcb39SKevin Wolf s->target = NULL; 6574ef85a9cSKevin Wolf 6584ef85a9cSKevin Wolf /* We don't access the source any more. Dropping any WRITE/RESIZE is 6594ef85a9cSKevin Wolf * required before it could become a backing file of target_bs. */ 660f94dc3b4SMax Reitz bs_opaque->stop = true; 661f94dc3b4SMax Reitz bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, 6624ef85a9cSKevin Wolf &error_abort); 663737efc1eSJohn Snow if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { 6644ef85a9cSKevin Wolf BlockDriverState *backing = s->is_none_mode ? src : s->base; 6654ef85a9cSKevin Wolf if (backing_bs(target_bs) != backing) { 66612fa4af6SKevin Wolf bdrv_set_backing_hd(target_bs, backing, &local_err); 66712fa4af6SKevin Wolf if (local_err) { 66812fa4af6SKevin Wolf error_report_err(local_err); 6697b508f6bSJohn Snow ret = -EPERM; 67012fa4af6SKevin Wolf } 6714ef85a9cSKevin Wolf } 6724ef85a9cSKevin Wolf } 6735a7e7a0bSStefan Hajnoczi 6745a7e7a0bSStefan Hajnoczi if (s->to_replace) { 6755a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 6765a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 6775a7e7a0bSStefan Hajnoczi } 6785a7e7a0bSStefan Hajnoczi 679737efc1eSJohn Snow if (s->should_complete && !abort) { 680737efc1eSJohn Snow BlockDriverState *to_replace = s->to_replace ?: src; 6811ba79388SAlberto Garcia bool ro = bdrv_is_read_only(to_replace); 68240365552SKevin Wolf 6831ba79388SAlberto Garcia if (ro != bdrv_is_read_only(target_bs)) { 6841ba79388SAlberto Garcia bdrv_reopen_set_read_only(target_bs, ro, NULL); 6855a7e7a0bSStefan Hajnoczi } 686b8804815SKevin Wolf 687b8804815SKevin Wolf /* The mirror job has no requests in flight any more, but we need to 688b8804815SKevin Wolf * drain potential other users of the BDS before changing the graph. */ 6895e771752SSergio Lopez assert(s->in_drain); 690e253f4b8SKevin Wolf bdrv_drained_begin(target_bs); 6915fe31c25SKevin Wolf bdrv_replace_node(to_replace, target_bs, &local_err); 692e253f4b8SKevin Wolf bdrv_drained_end(target_bs); 6935fe31c25SKevin Wolf if (local_err) { 6945fe31c25SKevin Wolf error_report_err(local_err); 6957b508f6bSJohn Snow ret = -EPERM; 6965fe31c25SKevin Wolf } 6975a7e7a0bSStefan Hajnoczi } 6985a7e7a0bSStefan Hajnoczi if (s->to_replace) { 6995a7e7a0bSStefan Hajnoczi bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 7005a7e7a0bSStefan Hajnoczi error_free(s->replace_blocker); 7015a7e7a0bSStefan Hajnoczi bdrv_unref(s->to_replace); 7025a7e7a0bSStefan Hajnoczi } 7035a7e7a0bSStefan Hajnoczi if (replace_aio_context) { 7045a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 7055a7e7a0bSStefan Hajnoczi } 7065a7e7a0bSStefan Hajnoczi g_free(s->replaces); 7077d9fcb39SKevin Wolf bdrv_unref(target_bs); 7084ef85a9cSKevin Wolf 709f94dc3b4SMax Reitz /* 710f94dc3b4SMax Reitz * Remove the mirror filter driver from the graph. Before this, get rid of 7114ef85a9cSKevin Wolf * the blockers on the intermediate nodes so that the resulting state is 712f94dc3b4SMax Reitz * valid. 713f94dc3b4SMax Reitz */ 7141908a559SKevin Wolf block_job_remove_all_bdrv(bjob); 7155fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 7164ef85a9cSKevin Wolf 7174ef85a9cSKevin Wolf /* We just changed the BDS the job BB refers to (with either or both of the 7185fe31c25SKevin Wolf * bdrv_replace_node() calls), so switch the BB back so the cleanup does 7195fe31c25SKevin Wolf * the right thing. We don't need any permissions any more now. */ 7201908a559SKevin Wolf blk_remove_bs(bjob->blk); 7211908a559SKevin Wolf blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); 7221908a559SKevin Wolf blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort); 7234ef85a9cSKevin Wolf 724429076e8SMax Reitz bs_opaque->job = NULL; 7254ef85a9cSKevin Wolf 726176c3699SFam Zheng bdrv_drained_end(src); 7275e771752SSergio Lopez s->in_drain = false; 7284ef85a9cSKevin Wolf bdrv_unref(mirror_top_bs); 7293f09bfbcSKevin Wolf bdrv_unref(src); 7307b508f6bSJohn Snow 731737efc1eSJohn Snow return ret; 732737efc1eSJohn Snow } 733737efc1eSJohn Snow 734737efc1eSJohn Snow static int mirror_prepare(Job *job) 735737efc1eSJohn Snow { 736737efc1eSJohn Snow return mirror_exit_common(job); 737737efc1eSJohn Snow } 738737efc1eSJohn Snow 739737efc1eSJohn Snow static void mirror_abort(Job *job) 740737efc1eSJohn Snow { 741737efc1eSJohn Snow int ret = mirror_exit_common(job); 742737efc1eSJohn Snow assert(ret == 0); 7435a7e7a0bSStefan Hajnoczi } 7445a7e7a0bSStefan Hajnoczi 745537c3d4fSStefan Hajnoczi static void coroutine_fn mirror_throttle(MirrorBlockJob *s) 74649efb1f5SDenis V. Lunev { 74749efb1f5SDenis V. Lunev int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 74849efb1f5SDenis V. Lunev 74918bb6928SKevin Wolf if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { 75049efb1f5SDenis V. Lunev s->last_pause_ns = now; 7515d43e86eSKevin Wolf job_sleep_ns(&s->common.job, 0); 75249efb1f5SDenis V. Lunev } else { 753da01ff7fSKevin Wolf job_pause_point(&s->common.job); 75449efb1f5SDenis V. Lunev } 75549efb1f5SDenis V. Lunev } 75649efb1f5SDenis V. Lunev 757c0b363adSDenis V. Lunev static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) 758c0b363adSDenis V. Lunev { 75923ca459aSEric Blake int64_t offset; 760c0b363adSDenis V. Lunev BlockDriverState *base = s->base; 761138f9fffSMax Reitz BlockDriverState *bs = s->mirror_top_bs->backing->bs; 762c0b363adSDenis V. Lunev BlockDriverState *target_bs = blk_bs(s->target); 76323ca459aSEric Blake int ret; 76451b0a488SEric Blake int64_t count; 765c0b363adSDenis V. Lunev 766b7d5062cSDenis V. Lunev if (base == NULL && !bdrv_has_zero_init(target_bs)) { 767c7c2769cSDenis V. Lunev if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { 768e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); 769b7d5062cSDenis V. Lunev return 0; 770b7d5062cSDenis V. Lunev } 771b7d5062cSDenis V. Lunev 77290ab48ebSAnton Nefedov s->initial_zeroing_ongoing = true; 77323ca459aSEric Blake for (offset = 0; offset < s->bdev_length; ) { 77423ca459aSEric Blake int bytes = MIN(s->bdev_length - offset, 77523ca459aSEric Blake QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 776c7c2769cSDenis V. Lunev 777c7c2769cSDenis V. Lunev mirror_throttle(s); 778c7c2769cSDenis V. Lunev 779daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job)) { 78090ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 781c7c2769cSDenis V. Lunev return 0; 782c7c2769cSDenis V. Lunev } 783c7c2769cSDenis V. Lunev 784c7c2769cSDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT) { 78567adf4b3SEric Blake trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, 78667adf4b3SEric Blake s->in_flight); 7871181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 788c7c2769cSDenis V. Lunev continue; 789c7c2769cSDenis V. Lunev } 790c7c2769cSDenis V. Lunev 7914295c5fcSMax Reitz mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); 79223ca459aSEric Blake offset += bytes; 793c7c2769cSDenis V. Lunev } 794c7c2769cSDenis V. Lunev 795bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 79690ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 797c7c2769cSDenis V. Lunev } 798c7c2769cSDenis V. Lunev 799c0b363adSDenis V. Lunev /* First part, loop on the sectors and initialize the dirty bitmap. */ 80023ca459aSEric Blake for (offset = 0; offset < s->bdev_length; ) { 801c0b363adSDenis V. Lunev /* Just to make sure we are not exceeding int limit. */ 80223ca459aSEric Blake int bytes = MIN(s->bdev_length - offset, 80323ca459aSEric Blake QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 804c0b363adSDenis V. Lunev 805c0b363adSDenis V. Lunev mirror_throttle(s); 806c0b363adSDenis V. Lunev 807daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job)) { 808c0b363adSDenis V. Lunev return 0; 809c0b363adSDenis V. Lunev } 810c0b363adSDenis V. Lunev 811*170d3bd3SAndrey Shinkevich ret = bdrv_is_allocated_above(bs, base, false, offset, bytes, &count); 812c0b363adSDenis V. Lunev if (ret < 0) { 813c0b363adSDenis V. Lunev return ret; 814c0b363adSDenis V. Lunev } 815c0b363adSDenis V. Lunev 81623ca459aSEric Blake assert(count); 817b7d5062cSDenis V. Lunev if (ret == 1) { 81823ca459aSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); 819c0b363adSDenis V. Lunev } 82023ca459aSEric Blake offset += count; 821c0b363adSDenis V. Lunev } 822c0b363adSDenis V. Lunev return 0; 823c0b363adSDenis V. Lunev } 824c0b363adSDenis V. Lunev 825bdffb31dSPaolo Bonzini /* Called when going out of the streaming phase to flush the bulk of the 826bdffb31dSPaolo Bonzini * data to the medium, or just before completing. 827bdffb31dSPaolo Bonzini */ 828bdffb31dSPaolo Bonzini static int mirror_flush(MirrorBlockJob *s) 829bdffb31dSPaolo Bonzini { 830bdffb31dSPaolo Bonzini int ret = blk_flush(s->target); 831bdffb31dSPaolo Bonzini if (ret < 0) { 832bdffb31dSPaolo Bonzini if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { 833bdffb31dSPaolo Bonzini s->ret = ret; 834bdffb31dSPaolo Bonzini } 835bdffb31dSPaolo Bonzini } 836bdffb31dSPaolo Bonzini return ret; 837bdffb31dSPaolo Bonzini } 838bdffb31dSPaolo Bonzini 839f67432a2SJohn Snow static int coroutine_fn mirror_run(Job *job, Error **errp) 840893f7ebaSPaolo Bonzini { 841f67432a2SJohn Snow MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 842138f9fffSMax Reitz BlockDriverState *bs = s->mirror_top_bs->backing->bs; 843e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 8449a0cec66SPaolo Bonzini bool need_drain = true; 845c0b363adSDenis V. Lunev int64_t length; 846b812f671SPaolo Bonzini BlockDriverInfo bdi; 8471d33936eSJeff Cody char backing_filename[2]; /* we only need 2 characters because we are only 8481d33936eSJeff Cody checking for a NULL string */ 849893f7ebaSPaolo Bonzini int ret = 0; 850893f7ebaSPaolo Bonzini 851daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job)) { 852893f7ebaSPaolo Bonzini goto immediate_exit; 853893f7ebaSPaolo Bonzini } 854893f7ebaSPaolo Bonzini 855b21c7652SMax Reitz s->bdev_length = bdrv_getlength(bs); 856b21c7652SMax Reitz if (s->bdev_length < 0) { 857b21c7652SMax Reitz ret = s->bdev_length; 858373df5b1SFam Zheng goto immediate_exit; 859becc347eSKevin Wolf } 860becc347eSKevin Wolf 861becc347eSKevin Wolf /* Active commit must resize the base image if its size differs from the 862becc347eSKevin Wolf * active layer. */ 863becc347eSKevin Wolf if (s->base == blk_bs(s->target)) { 864becc347eSKevin Wolf int64_t base_length; 865becc347eSKevin Wolf 866becc347eSKevin Wolf base_length = blk_getlength(s->target); 867becc347eSKevin Wolf if (base_length < 0) { 868becc347eSKevin Wolf ret = base_length; 869becc347eSKevin Wolf goto immediate_exit; 870becc347eSKevin Wolf } 871becc347eSKevin Wolf 872becc347eSKevin Wolf if (s->bdev_length > base_length) { 8733a691c50SMax Reitz ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, 8743a691c50SMax Reitz NULL); 875becc347eSKevin Wolf if (ret < 0) { 876becc347eSKevin Wolf goto immediate_exit; 877becc347eSKevin Wolf } 878becc347eSKevin Wolf } 879becc347eSKevin Wolf } 880becc347eSKevin Wolf 881becc347eSKevin Wolf if (s->bdev_length == 0) { 8822e1795b5SKevin Wolf /* Transition to the READY state and wait for complete. */ 8832e1795b5SKevin Wolf job_transition_to_ready(&s->common.job); 8849e48b025SFam Zheng s->synced = true; 885d06107adSMax Reitz s->actively_synced = true; 886daa7f2f9SKevin Wolf while (!job_is_cancelled(&s->common.job) && !s->should_complete) { 887198c49ccSKevin Wolf job_yield(&s->common.job); 8889e48b025SFam Zheng } 889daa7f2f9SKevin Wolf s->common.job.cancelled = false; 8909e48b025SFam Zheng goto immediate_exit; 891893f7ebaSPaolo Bonzini } 892893f7ebaSPaolo Bonzini 893b21c7652SMax Reitz length = DIV_ROUND_UP(s->bdev_length, s->granularity); 894402a4741SPaolo Bonzini s->in_flight_bitmap = bitmap_new(length); 895402a4741SPaolo Bonzini 896b812f671SPaolo Bonzini /* If we have no backing file yet in the destination, we cannot let 897b812f671SPaolo Bonzini * the destination do COW. Instead, we copy sectors around the 898b812f671SPaolo Bonzini * dirty data if needed. We need a bitmap to do that. 899b812f671SPaolo Bonzini */ 900e253f4b8SKevin Wolf bdrv_get_backing_filename(target_bs, backing_filename, 901b812f671SPaolo Bonzini sizeof(backing_filename)); 902e253f4b8SKevin Wolf if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { 903b436982fSEric Blake s->target_cluster_size = bdi.cluster_size; 904b436982fSEric Blake } else { 905b436982fSEric Blake s->target_cluster_size = BDRV_SECTOR_SIZE; 906c3cc95bdSFam Zheng } 907b436982fSEric Blake if (backing_filename[0] && !target_bs->backing && 908b436982fSEric Blake s->granularity < s->target_cluster_size) { 909b436982fSEric Blake s->buf_size = MAX(s->buf_size, s->target_cluster_size); 910b812f671SPaolo Bonzini s->cow_bitmap = bitmap_new(length); 911b812f671SPaolo Bonzini } 912e253f4b8SKevin Wolf s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); 913b812f671SPaolo Bonzini 9147504edf4SKevin Wolf s->buf = qemu_try_blockalign(bs, s->buf_size); 9157504edf4SKevin Wolf if (s->buf == NULL) { 9167504edf4SKevin Wolf ret = -ENOMEM; 9177504edf4SKevin Wolf goto immediate_exit; 9187504edf4SKevin Wolf } 9197504edf4SKevin Wolf 920402a4741SPaolo Bonzini mirror_free_init(s); 921893f7ebaSPaolo Bonzini 92249efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 92303544a6eSFam Zheng if (!s->is_none_mode) { 924c0b363adSDenis V. Lunev ret = mirror_dirty_init(s); 925daa7f2f9SKevin Wolf if (ret < 0 || job_is_cancelled(&s->common.job)) { 9264c0cbd6fSFam Zheng goto immediate_exit; 9274c0cbd6fSFam Zheng } 928893f7ebaSPaolo Bonzini } 929893f7ebaSPaolo Bonzini 930dc162c8eSFam Zheng assert(!s->dbi); 931715a74d8SEric Blake s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); 932893f7ebaSPaolo Bonzini for (;;) { 933cc8c9d6cSPaolo Bonzini uint64_t delay_ns = 0; 93449efb1f5SDenis V. Lunev int64_t cnt, delta; 935893f7ebaSPaolo Bonzini bool should_complete; 936893f7ebaSPaolo Bonzini 937d06107adSMax Reitz /* Do not start passive operations while there are active 938d06107adSMax Reitz * writes in progress */ 939d06107adSMax Reitz while (s->in_active_write_counter) { 940d06107adSMax Reitz mirror_wait_for_any_operation(s, true); 941d06107adSMax Reitz } 942d06107adSMax Reitz 943bd48bde8SPaolo Bonzini if (s->ret < 0) { 944bd48bde8SPaolo Bonzini ret = s->ret; 945893f7ebaSPaolo Bonzini goto immediate_exit; 946893f7ebaSPaolo Bonzini } 947bd48bde8SPaolo Bonzini 948da01ff7fSKevin Wolf job_pause_point(&s->common.job); 949565ac01fSStefan Hajnoczi 95020dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 95105df8a6aSKevin Wolf /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is 95205df8a6aSKevin Wolf * the number of bytes currently being processed; together those are 95305df8a6aSKevin Wolf * the current remaining operation length */ 95430a5c887SKevin Wolf job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt); 955bd48bde8SPaolo Bonzini 956bd48bde8SPaolo Bonzini /* Note that even when no rate limit is applied we need to yield 957a7282330SFam Zheng * periodically with no pending I/O so that bdrv_drain_all() returns. 95818bb6928SKevin Wolf * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is 95918bb6928SKevin Wolf * an error, or when the source is clean, whichever comes first. */ 96049efb1f5SDenis V. Lunev delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; 96118bb6928SKevin Wolf if (delta < BLOCK_JOB_SLICE_TIME && 962bd48bde8SPaolo Bonzini s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 963cf56a3c6SDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || 964402a4741SPaolo Bonzini (cnt == 0 && s->in_flight > 0)) { 9659a46dba7SEric Blake trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); 9661181e19aSMax Reitz mirror_wait_for_free_in_flight_slot(s); 967bd48bde8SPaolo Bonzini continue; 968bd48bde8SPaolo Bonzini } else if (cnt != 0) { 969cc8c9d6cSPaolo Bonzini delay_ns = mirror_iteration(s); 970893f7ebaSPaolo Bonzini } 971cc8c9d6cSPaolo Bonzini } 972893f7ebaSPaolo Bonzini 973893f7ebaSPaolo Bonzini should_complete = false; 974bd48bde8SPaolo Bonzini if (s->in_flight == 0 && cnt == 0) { 975893f7ebaSPaolo Bonzini trace_mirror_before_flush(s); 976bdffb31dSPaolo Bonzini if (!s->synced) { 977bdffb31dSPaolo Bonzini if (mirror_flush(s) < 0) { 978bdffb31dSPaolo Bonzini /* Go check s->ret. */ 979bdffb31dSPaolo Bonzini continue; 980893f7ebaSPaolo Bonzini } 981893f7ebaSPaolo Bonzini /* We're out of the streaming phase. From now on, if the job 982893f7ebaSPaolo Bonzini * is cancelled we will actually complete all pending I/O and 983893f7ebaSPaolo Bonzini * report completion. This way, block-job-cancel will leave 984893f7ebaSPaolo Bonzini * the target in a consistent state. 985893f7ebaSPaolo Bonzini */ 9862e1795b5SKevin Wolf job_transition_to_ready(&s->common.job); 987d63ffd87SPaolo Bonzini s->synced = true; 988d06107adSMax Reitz if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) { 989d06107adSMax Reitz s->actively_synced = true; 990d06107adSMax Reitz } 991d63ffd87SPaolo Bonzini } 992d63ffd87SPaolo Bonzini 993d63ffd87SPaolo Bonzini should_complete = s->should_complete || 994daa7f2f9SKevin Wolf job_is_cancelled(&s->common.job); 99520dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 996893f7ebaSPaolo Bonzini } 997893f7ebaSPaolo Bonzini 998893f7ebaSPaolo Bonzini if (cnt == 0 && should_complete) { 999893f7ebaSPaolo Bonzini /* The dirty bitmap is not updated while operations are pending. 1000893f7ebaSPaolo Bonzini * If we're about to exit, wait for pending operations before 1001893f7ebaSPaolo Bonzini * calling bdrv_get_dirty_count(bs), or we may exit while the 1002893f7ebaSPaolo Bonzini * source has dirty data to copy! 1003893f7ebaSPaolo Bonzini * 1004893f7ebaSPaolo Bonzini * Note that I/O can be submitted by the guest while 10059a0cec66SPaolo Bonzini * mirror_populate runs, so pause it now. Before deciding 10069a0cec66SPaolo Bonzini * whether to switch to target check one last time if I/O has 10079a0cec66SPaolo Bonzini * come in the meanwhile, and if not flush the data to disk. 1008893f7ebaSPaolo Bonzini */ 10099a46dba7SEric Blake trace_mirror_before_drain(s, cnt); 10109a0cec66SPaolo Bonzini 10115e771752SSergio Lopez s->in_drain = true; 10129a0cec66SPaolo Bonzini bdrv_drained_begin(bs); 101320dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 1014bdffb31dSPaolo Bonzini if (cnt > 0 || mirror_flush(s) < 0) { 10159a0cec66SPaolo Bonzini bdrv_drained_end(bs); 10165e771752SSergio Lopez s->in_drain = false; 10179a0cec66SPaolo Bonzini continue; 10189a0cec66SPaolo Bonzini } 10199a0cec66SPaolo Bonzini 10209a0cec66SPaolo Bonzini /* The two disks are in sync. Exit and report successful 10219a0cec66SPaolo Bonzini * completion. 10229a0cec66SPaolo Bonzini */ 10239a0cec66SPaolo Bonzini assert(QLIST_EMPTY(&bs->tracked_requests)); 1024daa7f2f9SKevin Wolf s->common.job.cancelled = false; 10259a0cec66SPaolo Bonzini need_drain = false; 10269a0cec66SPaolo Bonzini break; 1027893f7ebaSPaolo Bonzini } 1028893f7ebaSPaolo Bonzini 1029893f7ebaSPaolo Bonzini ret = 0; 1030ddc4115eSStefan Hajnoczi 1031ddc4115eSStefan Hajnoczi if (s->synced && !should_complete) { 103218bb6928SKevin Wolf delay_ns = (s->in_flight == 0 && 103318bb6928SKevin Wolf cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0); 1034ddc4115eSStefan Hajnoczi } 10359a46dba7SEric Blake trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); 10365d43e86eSKevin Wolf job_sleep_ns(&s->common.job, delay_ns); 1037daa7f2f9SKevin Wolf if (job_is_cancelled(&s->common.job) && 1038004e95dfSKevin Wolf (!s->synced || s->common.job.force_cancel)) 1039eb36639fSMax Reitz { 1040893f7ebaSPaolo Bonzini break; 1041893f7ebaSPaolo Bonzini } 104249efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1043893f7ebaSPaolo Bonzini } 1044893f7ebaSPaolo Bonzini 1045893f7ebaSPaolo Bonzini immediate_exit: 1046bd48bde8SPaolo Bonzini if (s->in_flight > 0) { 1047bd48bde8SPaolo Bonzini /* We get here only if something went wrong. Either the job failed, 1048bd48bde8SPaolo Bonzini * or it was cancelled prematurely so that we do not guarantee that 1049bd48bde8SPaolo Bonzini * the target is a copy of the source. 1050bd48bde8SPaolo Bonzini */ 1051004e95dfSKevin Wolf assert(ret < 0 || ((s->common.job.force_cancel || !s->synced) && 1052daa7f2f9SKevin Wolf job_is_cancelled(&s->common.job))); 10539a0cec66SPaolo Bonzini assert(need_drain); 1054bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 1055bd48bde8SPaolo Bonzini } 1056bd48bde8SPaolo Bonzini 1057bd48bde8SPaolo Bonzini assert(s->in_flight == 0); 10587191bf31SMarkus Armbruster qemu_vfree(s->buf); 1059b812f671SPaolo Bonzini g_free(s->cow_bitmap); 1060402a4741SPaolo Bonzini g_free(s->in_flight_bitmap); 1061dc162c8eSFam Zheng bdrv_dirty_iter_free(s->dbi); 10625a7e7a0bSStefan Hajnoczi 10639a0cec66SPaolo Bonzini if (need_drain) { 10645e771752SSergio Lopez s->in_drain = true; 1065e253f4b8SKevin Wolf bdrv_drained_begin(bs); 10669a0cec66SPaolo Bonzini } 1067f67432a2SJohn Snow 1068f67432a2SJohn Snow return ret; 1069893f7ebaSPaolo Bonzini } 1070893f7ebaSPaolo Bonzini 10713453d972SKevin Wolf static void mirror_complete(Job *job, Error **errp) 1072d63ffd87SPaolo Bonzini { 10733453d972SKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 10744ef85a9cSKevin Wolf BlockDriverState *target; 1075d63ffd87SPaolo Bonzini 1076274fcceeSMax Reitz target = blk_bs(s->target); 1077274fcceeSMax Reitz 1078d63ffd87SPaolo Bonzini if (!s->synced) { 10799df229c3SAlberto Garcia error_setg(errp, "The active block job '%s' cannot be completed", 10803453d972SKevin Wolf job->id); 1081d63ffd87SPaolo Bonzini return; 1082d63ffd87SPaolo Bonzini } 1083d63ffd87SPaolo Bonzini 1084274fcceeSMax Reitz if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { 1085274fcceeSMax Reitz int ret; 1086274fcceeSMax Reitz 1087274fcceeSMax Reitz assert(!target->backing); 1088274fcceeSMax Reitz ret = bdrv_open_backing_file(target, NULL, "backing", errp); 1089274fcceeSMax Reitz if (ret < 0) { 1090274fcceeSMax Reitz return; 1091274fcceeSMax Reitz } 1092274fcceeSMax Reitz } 1093274fcceeSMax Reitz 109415d67298SChanglong Xie /* block all operations on to_replace bs */ 109509158f00SBenoît Canet if (s->replaces) { 10965a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context; 10975a7e7a0bSStefan Hajnoczi 1098e12f3784SWen Congyang s->to_replace = bdrv_find_node(s->replaces); 109909158f00SBenoît Canet if (!s->to_replace) { 1100e12f3784SWen Congyang error_setg(errp, "Node name '%s' not found", s->replaces); 110109158f00SBenoît Canet return; 110209158f00SBenoît Canet } 110309158f00SBenoît Canet 11045a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 11055a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 11065a7e7a0bSStefan Hajnoczi 11074ef85a9cSKevin Wolf /* TODO Translate this into permission system. Current definition of 11084ef85a9cSKevin Wolf * GRAPH_MOD would require to request it for the parents; they might 11094ef85a9cSKevin Wolf * not even be BlockDriverStates, however, so a BdrvChild can't address 11104ef85a9cSKevin Wolf * them. May need redefinition of GRAPH_MOD. */ 111109158f00SBenoît Canet error_setg(&s->replace_blocker, 111209158f00SBenoît Canet "block device is in use by block-job-complete"); 111309158f00SBenoît Canet bdrv_op_block_all(s->to_replace, s->replace_blocker); 111409158f00SBenoît Canet bdrv_ref(s->to_replace); 11155a7e7a0bSStefan Hajnoczi 11165a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 111709158f00SBenoît Canet } 111809158f00SBenoît Canet 1119d63ffd87SPaolo Bonzini s->should_complete = true; 11203d70ff53SKevin Wolf job_enter(job); 1121d63ffd87SPaolo Bonzini } 1122d63ffd87SPaolo Bonzini 1123537c3d4fSStefan Hajnoczi static void coroutine_fn mirror_pause(Job *job) 1124565ac01fSStefan Hajnoczi { 1125da01ff7fSKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 1126565ac01fSStefan Hajnoczi 1127bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 1128565ac01fSStefan Hajnoczi } 1129565ac01fSStefan Hajnoczi 113089bd0305SKevin Wolf static bool mirror_drained_poll(BlockJob *job) 113189bd0305SKevin Wolf { 113289bd0305SKevin Wolf MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 11335e771752SSergio Lopez 11345e771752SSergio Lopez /* If the job isn't paused nor cancelled, we can't be sure that it won't 11355e771752SSergio Lopez * issue more requests. We make an exception if we've reached this point 11365e771752SSergio Lopez * from one of our own drain sections, to avoid a deadlock waiting for 11375e771752SSergio Lopez * ourselves. 11385e771752SSergio Lopez */ 11395e771752SSergio Lopez if (!s->common.job.paused && !s->common.job.cancelled && !s->in_drain) { 11405e771752SSergio Lopez return true; 11415e771752SSergio Lopez } 11425e771752SSergio Lopez 114389bd0305SKevin Wolf return !!s->in_flight; 114489bd0305SKevin Wolf } 114589bd0305SKevin Wolf 1146bae8196dSPaolo Bonzini static void mirror_drain(BlockJob *job) 1147bae8196dSPaolo Bonzini { 1148bae8196dSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 1149bae8196dSPaolo Bonzini 1150bae8196dSPaolo Bonzini /* Need to keep a reference in case blk_drain triggers execution 1151bae8196dSPaolo Bonzini * of mirror_complete... 1152bae8196dSPaolo Bonzini */ 1153bae8196dSPaolo Bonzini if (s->target) { 1154bae8196dSPaolo Bonzini BlockBackend *target = s->target; 1155bae8196dSPaolo Bonzini blk_ref(target); 1156bae8196dSPaolo Bonzini blk_drain(target); 1157bae8196dSPaolo Bonzini blk_unref(target); 1158bae8196dSPaolo Bonzini } 1159bae8196dSPaolo Bonzini } 1160bae8196dSPaolo Bonzini 11613fc4b10aSFam Zheng static const BlockJobDriver mirror_job_driver = { 116233e9e9bdSKevin Wolf .job_driver = { 1163893f7ebaSPaolo Bonzini .instance_size = sizeof(MirrorBlockJob), 11648e4c8700SKevin Wolf .job_type = JOB_TYPE_MIRROR, 116580fa2c75SKevin Wolf .free = block_job_free, 1166b15de828SKevin Wolf .user_resume = block_job_user_resume, 1167b69f777dSKevin Wolf .drain = block_job_drain, 1168f67432a2SJohn Snow .run = mirror_run, 1169737efc1eSJohn Snow .prepare = mirror_prepare, 1170737efc1eSJohn Snow .abort = mirror_abort, 1171565ac01fSStefan Hajnoczi .pause = mirror_pause, 1172da01ff7fSKevin Wolf .complete = mirror_complete, 11733453d972SKevin Wolf }, 117489bd0305SKevin Wolf .drained_poll = mirror_drained_poll, 1175bae8196dSPaolo Bonzini .drain = mirror_drain, 1176893f7ebaSPaolo Bonzini }; 1177893f7ebaSPaolo Bonzini 117803544a6eSFam Zheng static const BlockJobDriver commit_active_job_driver = { 117933e9e9bdSKevin Wolf .job_driver = { 118003544a6eSFam Zheng .instance_size = sizeof(MirrorBlockJob), 11818e4c8700SKevin Wolf .job_type = JOB_TYPE_COMMIT, 118280fa2c75SKevin Wolf .free = block_job_free, 1183b15de828SKevin Wolf .user_resume = block_job_user_resume, 1184b69f777dSKevin Wolf .drain = block_job_drain, 1185f67432a2SJohn Snow .run = mirror_run, 1186737efc1eSJohn Snow .prepare = mirror_prepare, 1187737efc1eSJohn Snow .abort = mirror_abort, 1188565ac01fSStefan Hajnoczi .pause = mirror_pause, 1189da01ff7fSKevin Wolf .complete = mirror_complete, 11903453d972SKevin Wolf }, 119189bd0305SKevin Wolf .drained_poll = mirror_drained_poll, 1192bae8196dSPaolo Bonzini .drain = mirror_drain, 119303544a6eSFam Zheng }; 119403544a6eSFam Zheng 1195537c3d4fSStefan Hajnoczi static void coroutine_fn 1196537c3d4fSStefan Hajnoczi do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, 1197d06107adSMax Reitz uint64_t offset, uint64_t bytes, 1198d06107adSMax Reitz QEMUIOVector *qiov, int flags) 1199d06107adSMax Reitz { 1200d06107adSMax Reitz QEMUIOVector target_qiov; 12011eaf1b0fSVladimir Sementsov-Ogievskiy uint64_t dirty_offset = offset; 12021eaf1b0fSVladimir Sementsov-Ogievskiy uint64_t dirty_bytes; 1203d06107adSMax Reitz 1204d06107adSMax Reitz if (qiov) { 1205d06107adSMax Reitz qemu_iovec_init(&target_qiov, qiov->niov); 1206d06107adSMax Reitz } 1207d06107adSMax Reitz 1208d06107adSMax Reitz while (true) { 1209d06107adSMax Reitz bool valid_area; 1210d06107adSMax Reitz int ret; 1211d06107adSMax Reitz 1212d06107adSMax Reitz bdrv_dirty_bitmap_lock(job->dirty_bitmap); 12131eaf1b0fSVladimir Sementsov-Ogievskiy dirty_bytes = MIN(offset + bytes - dirty_offset, INT_MAX); 12141eaf1b0fSVladimir Sementsov-Ogievskiy valid_area = bdrv_dirty_bitmap_next_dirty_area(job->dirty_bitmap, 12151eaf1b0fSVladimir Sementsov-Ogievskiy &dirty_offset, 12161eaf1b0fSVladimir Sementsov-Ogievskiy &dirty_bytes); 1217d06107adSMax Reitz if (!valid_area) { 1218d06107adSMax Reitz bdrv_dirty_bitmap_unlock(job->dirty_bitmap); 1219d06107adSMax Reitz break; 1220d06107adSMax Reitz } 1221d06107adSMax Reitz 1222d06107adSMax Reitz bdrv_reset_dirty_bitmap_locked(job->dirty_bitmap, 1223d06107adSMax Reitz dirty_offset, dirty_bytes); 1224d06107adSMax Reitz bdrv_dirty_bitmap_unlock(job->dirty_bitmap); 1225d06107adSMax Reitz 1226d06107adSMax Reitz job_progress_increase_remaining(&job->common.job, dirty_bytes); 1227d06107adSMax Reitz 1228d06107adSMax Reitz assert(dirty_offset - offset <= SIZE_MAX); 1229d06107adSMax Reitz if (qiov) { 1230d06107adSMax Reitz qemu_iovec_reset(&target_qiov); 1231d06107adSMax Reitz qemu_iovec_concat(&target_qiov, qiov, 1232d06107adSMax Reitz dirty_offset - offset, dirty_bytes); 1233d06107adSMax Reitz } 1234d06107adSMax Reitz 1235d06107adSMax Reitz switch (method) { 1236d06107adSMax Reitz case MIRROR_METHOD_COPY: 1237d06107adSMax Reitz ret = blk_co_pwritev(job->target, dirty_offset, dirty_bytes, 1238d06107adSMax Reitz qiov ? &target_qiov : NULL, flags); 1239d06107adSMax Reitz break; 1240d06107adSMax Reitz 1241d06107adSMax Reitz case MIRROR_METHOD_ZERO: 1242d06107adSMax Reitz assert(!qiov); 1243d06107adSMax Reitz ret = blk_co_pwrite_zeroes(job->target, dirty_offset, dirty_bytes, 1244d06107adSMax Reitz flags); 1245d06107adSMax Reitz break; 1246d06107adSMax Reitz 1247d06107adSMax Reitz case MIRROR_METHOD_DISCARD: 1248d06107adSMax Reitz assert(!qiov); 1249d06107adSMax Reitz ret = blk_co_pdiscard(job->target, dirty_offset, dirty_bytes); 1250d06107adSMax Reitz break; 1251d06107adSMax Reitz 1252d06107adSMax Reitz default: 1253d06107adSMax Reitz abort(); 1254d06107adSMax Reitz } 1255d06107adSMax Reitz 1256d06107adSMax Reitz if (ret >= 0) { 1257d06107adSMax Reitz job_progress_update(&job->common.job, dirty_bytes); 1258d06107adSMax Reitz } else { 1259d06107adSMax Reitz BlockErrorAction action; 1260d06107adSMax Reitz 1261d06107adSMax Reitz bdrv_set_dirty_bitmap(job->dirty_bitmap, dirty_offset, dirty_bytes); 1262d06107adSMax Reitz job->actively_synced = false; 1263d06107adSMax Reitz 1264d06107adSMax Reitz action = mirror_error_action(job, false, -ret); 1265d06107adSMax Reitz if (action == BLOCK_ERROR_ACTION_REPORT) { 1266d06107adSMax Reitz if (!job->ret) { 1267d06107adSMax Reitz job->ret = ret; 1268d06107adSMax Reitz } 1269d06107adSMax Reitz break; 1270d06107adSMax Reitz } 1271d06107adSMax Reitz } 12721eaf1b0fSVladimir Sementsov-Ogievskiy 12731eaf1b0fSVladimir Sementsov-Ogievskiy dirty_offset += dirty_bytes; 1274d06107adSMax Reitz } 1275d06107adSMax Reitz 1276d06107adSMax Reitz if (qiov) { 1277d06107adSMax Reitz qemu_iovec_destroy(&target_qiov); 1278d06107adSMax Reitz } 1279d06107adSMax Reitz } 1280d06107adSMax Reitz 1281d06107adSMax Reitz static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, 1282d06107adSMax Reitz uint64_t offset, 1283d06107adSMax Reitz uint64_t bytes) 1284d06107adSMax Reitz { 1285d06107adSMax Reitz MirrorOp *op; 1286d06107adSMax Reitz uint64_t start_chunk = offset / s->granularity; 1287d06107adSMax Reitz uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); 1288d06107adSMax Reitz 1289d06107adSMax Reitz op = g_new(MirrorOp, 1); 1290d06107adSMax Reitz *op = (MirrorOp){ 1291d06107adSMax Reitz .s = s, 1292d06107adSMax Reitz .offset = offset, 1293d06107adSMax Reitz .bytes = bytes, 1294d06107adSMax Reitz .is_active_write = true, 1295d06107adSMax Reitz }; 1296d06107adSMax Reitz qemu_co_queue_init(&op->waiting_requests); 1297d06107adSMax Reitz QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); 1298d06107adSMax Reitz 1299d06107adSMax Reitz s->in_active_write_counter++; 1300d06107adSMax Reitz 1301d06107adSMax Reitz mirror_wait_on_conflicts(op, s, offset, bytes); 1302d06107adSMax Reitz 1303d06107adSMax Reitz bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); 1304d06107adSMax Reitz 1305d06107adSMax Reitz return op; 1306d06107adSMax Reitz } 1307d06107adSMax Reitz 1308d06107adSMax Reitz static void coroutine_fn active_write_settle(MirrorOp *op) 1309d06107adSMax Reitz { 1310d06107adSMax Reitz uint64_t start_chunk = op->offset / op->s->granularity; 1311d06107adSMax Reitz uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, 1312d06107adSMax Reitz op->s->granularity); 1313d06107adSMax Reitz 1314d06107adSMax Reitz if (!--op->s->in_active_write_counter && op->s->actively_synced) { 1315d06107adSMax Reitz BdrvChild *source = op->s->mirror_top_bs->backing; 1316d06107adSMax Reitz 1317d06107adSMax Reitz if (QLIST_FIRST(&source->bs->parents) == source && 1318d06107adSMax Reitz QLIST_NEXT(source, next_parent) == NULL) 1319d06107adSMax Reitz { 1320d06107adSMax Reitz /* Assert that we are back in sync once all active write 1321d06107adSMax Reitz * operations are settled. 1322d06107adSMax Reitz * Note that we can only assert this if the mirror node 1323d06107adSMax Reitz * is the source node's only parent. */ 1324d06107adSMax Reitz assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); 1325d06107adSMax Reitz } 1326d06107adSMax Reitz } 1327d06107adSMax Reitz bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); 1328d06107adSMax Reitz QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); 1329d06107adSMax Reitz qemu_co_queue_restart_all(&op->waiting_requests); 1330d06107adSMax Reitz g_free(op); 1331d06107adSMax Reitz } 1332d06107adSMax Reitz 13334ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, 13344ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 13354ef85a9cSKevin Wolf { 13364ef85a9cSKevin Wolf return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); 13374ef85a9cSKevin Wolf } 13384ef85a9cSKevin Wolf 1339d06107adSMax Reitz static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs, 1340d06107adSMax Reitz MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, 1341d06107adSMax Reitz int flags) 1342d06107adSMax Reitz { 1343d06107adSMax Reitz MirrorOp *op = NULL; 1344d06107adSMax Reitz MirrorBDSOpaque *s = bs->opaque; 1345d06107adSMax Reitz int ret = 0; 1346d06107adSMax Reitz bool copy_to_target; 1347d06107adSMax Reitz 1348d06107adSMax Reitz copy_to_target = s->job->ret >= 0 && 1349d06107adSMax Reitz s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; 1350d06107adSMax Reitz 1351d06107adSMax Reitz if (copy_to_target) { 1352d06107adSMax Reitz op = active_write_prepare(s->job, offset, bytes); 1353d06107adSMax Reitz } 1354d06107adSMax Reitz 1355d06107adSMax Reitz switch (method) { 1356d06107adSMax Reitz case MIRROR_METHOD_COPY: 1357d06107adSMax Reitz ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); 1358d06107adSMax Reitz break; 1359d06107adSMax Reitz 1360d06107adSMax Reitz case MIRROR_METHOD_ZERO: 1361d06107adSMax Reitz ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); 1362d06107adSMax Reitz break; 1363d06107adSMax Reitz 1364d06107adSMax Reitz case MIRROR_METHOD_DISCARD: 13650b9fd3f4SFam Zheng ret = bdrv_co_pdiscard(bs->backing, offset, bytes); 1366d06107adSMax Reitz break; 1367d06107adSMax Reitz 1368d06107adSMax Reitz default: 1369d06107adSMax Reitz abort(); 1370d06107adSMax Reitz } 1371d06107adSMax Reitz 1372d06107adSMax Reitz if (ret < 0) { 1373d06107adSMax Reitz goto out; 1374d06107adSMax Reitz } 1375d06107adSMax Reitz 1376d06107adSMax Reitz if (copy_to_target) { 1377d06107adSMax Reitz do_sync_target_write(s->job, method, offset, bytes, qiov, flags); 1378d06107adSMax Reitz } 1379d06107adSMax Reitz 1380d06107adSMax Reitz out: 1381d06107adSMax Reitz if (copy_to_target) { 1382d06107adSMax Reitz active_write_settle(op); 1383d06107adSMax Reitz } 1384d06107adSMax Reitz return ret; 1385d06107adSMax Reitz } 1386d06107adSMax Reitz 13874ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, 13884ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 13894ef85a9cSKevin Wolf { 1390d06107adSMax Reitz MirrorBDSOpaque *s = bs->opaque; 1391d06107adSMax Reitz QEMUIOVector bounce_qiov; 1392d06107adSMax Reitz void *bounce_buf; 1393d06107adSMax Reitz int ret = 0; 1394d06107adSMax Reitz bool copy_to_target; 1395d06107adSMax Reitz 1396d06107adSMax Reitz copy_to_target = s->job->ret >= 0 && 1397d06107adSMax Reitz s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; 1398d06107adSMax Reitz 1399d06107adSMax Reitz if (copy_to_target) { 1400d06107adSMax Reitz /* The guest might concurrently modify the data to write; but 1401d06107adSMax Reitz * the data on source and destination must match, so we have 1402d06107adSMax Reitz * to use a bounce buffer if we are going to write to the 1403d06107adSMax Reitz * target now. */ 1404d06107adSMax Reitz bounce_buf = qemu_blockalign(bs, bytes); 1405d06107adSMax Reitz iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); 1406d06107adSMax Reitz 1407d06107adSMax Reitz qemu_iovec_init(&bounce_qiov, 1); 1408d06107adSMax Reitz qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); 1409d06107adSMax Reitz qiov = &bounce_qiov; 1410d06107adSMax Reitz } 1411d06107adSMax Reitz 1412d06107adSMax Reitz ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov, 1413d06107adSMax Reitz flags); 1414d06107adSMax Reitz 1415d06107adSMax Reitz if (copy_to_target) { 1416d06107adSMax Reitz qemu_iovec_destroy(&bounce_qiov); 1417d06107adSMax Reitz qemu_vfree(bounce_buf); 1418d06107adSMax Reitz } 1419d06107adSMax Reitz 1420d06107adSMax Reitz return ret; 14214ef85a9cSKevin Wolf } 14224ef85a9cSKevin Wolf 14234ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) 14244ef85a9cSKevin Wolf { 1425ce960aa9SVladimir Sementsov-Ogievskiy if (bs->backing == NULL) { 1426ce960aa9SVladimir Sementsov-Ogievskiy /* we can be here after failed bdrv_append in mirror_start_job */ 1427ce960aa9SVladimir Sementsov-Ogievskiy return 0; 1428ce960aa9SVladimir Sementsov-Ogievskiy } 14294ef85a9cSKevin Wolf return bdrv_co_flush(bs->backing->bs); 14304ef85a9cSKevin Wolf } 14314ef85a9cSKevin Wolf 14324ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, 1433f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 14344ef85a9cSKevin Wolf { 1435d06107adSMax Reitz return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL, 1436d06107adSMax Reitz flags); 14374ef85a9cSKevin Wolf } 14384ef85a9cSKevin Wolf 14394ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, 1440f5a5ca79SManos Pitsidianakis int64_t offset, int bytes) 14414ef85a9cSKevin Wolf { 1442d06107adSMax Reitz return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes, 1443d06107adSMax Reitz NULL, 0); 14444ef85a9cSKevin Wolf } 14454ef85a9cSKevin Wolf 1446998b3a1eSMax Reitz static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs) 1447fd4a6493SKevin Wolf { 144818775ff3SVladimir Sementsov-Ogievskiy if (bs->backing == NULL) { 144918775ff3SVladimir Sementsov-Ogievskiy /* we can be here after failed bdrv_attach_child in 145018775ff3SVladimir Sementsov-Ogievskiy * bdrv_set_backing_hd */ 145118775ff3SVladimir Sementsov-Ogievskiy return; 145218775ff3SVladimir Sementsov-Ogievskiy } 1453fd4a6493SKevin Wolf pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), 1454fd4a6493SKevin Wolf bs->backing->bs->filename); 1455fd4a6493SKevin Wolf } 1456fd4a6493SKevin Wolf 14574ef85a9cSKevin Wolf static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, 14584ef85a9cSKevin Wolf const BdrvChildRole *role, 1459e0995dc3SKevin Wolf BlockReopenQueue *reopen_queue, 14604ef85a9cSKevin Wolf uint64_t perm, uint64_t shared, 14614ef85a9cSKevin Wolf uint64_t *nperm, uint64_t *nshared) 14624ef85a9cSKevin Wolf { 1463f94dc3b4SMax Reitz MirrorBDSOpaque *s = bs->opaque; 1464f94dc3b4SMax Reitz 1465f94dc3b4SMax Reitz if (s->stop) { 1466f94dc3b4SMax Reitz /* 1467f94dc3b4SMax Reitz * If the job is to be stopped, we do not need to forward 1468f94dc3b4SMax Reitz * anything to the real image. 1469f94dc3b4SMax Reitz */ 1470f94dc3b4SMax Reitz *nperm = 0; 1471f94dc3b4SMax Reitz *nshared = BLK_PERM_ALL; 1472f94dc3b4SMax Reitz return; 1473f94dc3b4SMax Reitz } 1474f94dc3b4SMax Reitz 14754ef85a9cSKevin Wolf /* Must be able to forward guest writes to the real image */ 14764ef85a9cSKevin Wolf *nperm = 0; 14774ef85a9cSKevin Wolf if (perm & BLK_PERM_WRITE) { 14784ef85a9cSKevin Wolf *nperm |= BLK_PERM_WRITE; 14794ef85a9cSKevin Wolf } 14804ef85a9cSKevin Wolf 14814ef85a9cSKevin Wolf *nshared = BLK_PERM_ALL; 14824ef85a9cSKevin Wolf } 14834ef85a9cSKevin Wolf 14844ef85a9cSKevin Wolf /* Dummy node that provides consistent read to its users without requiring it 14854ef85a9cSKevin Wolf * from its backing file and that allows writes on the backing file chain. */ 14864ef85a9cSKevin Wolf static BlockDriver bdrv_mirror_top = { 14874ef85a9cSKevin Wolf .format_name = "mirror_top", 14884ef85a9cSKevin Wolf .bdrv_co_preadv = bdrv_mirror_top_preadv, 14894ef85a9cSKevin Wolf .bdrv_co_pwritev = bdrv_mirror_top_pwritev, 14904ef85a9cSKevin Wolf .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, 14914ef85a9cSKevin Wolf .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, 14924ef85a9cSKevin Wolf .bdrv_co_flush = bdrv_mirror_top_flush, 14933e4d0e72SEric Blake .bdrv_co_block_status = bdrv_co_block_status_from_backing, 1494fd4a6493SKevin Wolf .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, 14954ef85a9cSKevin Wolf .bdrv_child_perm = bdrv_mirror_top_child_perm, 14964ef85a9cSKevin Wolf }; 14974ef85a9cSKevin Wolf 1498cc19f177SVladimir Sementsov-Ogievskiy static BlockJob *mirror_start_job( 1499cc19f177SVladimir Sementsov-Ogievskiy const char *job_id, BlockDriverState *bs, 150047970dfbSJohn Snow int creation_flags, BlockDriverState *target, 150147970dfbSJohn Snow const char *replaces, int64_t speed, 150247970dfbSJohn Snow uint32_t granularity, int64_t buf_size, 1503274fcceeSMax Reitz BlockMirrorBackingMode backing_mode, 150403544a6eSFam Zheng BlockdevOnError on_source_error, 1505b952b558SPaolo Bonzini BlockdevOnError on_target_error, 15060fc9f8eaSFam Zheng bool unmap, 1507097310b5SMarkus Armbruster BlockCompletionFunc *cb, 150851ccfa2dSFam Zheng void *opaque, 150903544a6eSFam Zheng const BlockJobDriver *driver, 1510b49f7eadSWen Congyang bool is_none_mode, BlockDriverState *base, 151151ccfa2dSFam Zheng bool auto_complete, const char *filter_node_name, 1512481debaaSMax Reitz bool is_mirror, MirrorCopyMode copy_mode, 151351ccfa2dSFam Zheng Error **errp) 1514893f7ebaSPaolo Bonzini { 1515893f7ebaSPaolo Bonzini MirrorBlockJob *s; 1516429076e8SMax Reitz MirrorBDSOpaque *bs_opaque; 15174ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 15184ef85a9cSKevin Wolf bool target_graph_mod; 15194ef85a9cSKevin Wolf bool target_is_backing; 1520b2c2832cSKevin Wolf Error *local_err = NULL; 1521d7086422SKevin Wolf int ret; 1522893f7ebaSPaolo Bonzini 1523eee13dfeSPaolo Bonzini if (granularity == 0) { 1524341ebc2fSJohn Snow granularity = bdrv_get_default_bitmap_granularity(target); 1525eee13dfeSPaolo Bonzini } 1526eee13dfeSPaolo Bonzini 152731826642SEric Blake assert(is_power_of_2(granularity)); 1528eee13dfeSPaolo Bonzini 152948ac0a4dSWen Congyang if (buf_size < 0) { 153048ac0a4dSWen Congyang error_setg(errp, "Invalid parameter 'buf-size'"); 1531cc19f177SVladimir Sementsov-Ogievskiy return NULL; 153248ac0a4dSWen Congyang } 153348ac0a4dSWen Congyang 153448ac0a4dSWen Congyang if (buf_size == 0) { 153548ac0a4dSWen Congyang buf_size = DEFAULT_MIRROR_BUF_SIZE; 153648ac0a4dSWen Congyang } 15375bc361b8SFam Zheng 153886fae10cSKevin Wolf if (bs == target) { 153986fae10cSKevin Wolf error_setg(errp, "Can't mirror node into itself"); 1540cc19f177SVladimir Sementsov-Ogievskiy return NULL; 154186fae10cSKevin Wolf } 154286fae10cSKevin Wolf 15434ef85a9cSKevin Wolf /* In the case of active commit, add dummy driver to provide consistent 15444ef85a9cSKevin Wolf * reads on the top, while disabling it in the intermediate nodes, and make 15454ef85a9cSKevin Wolf * the backing chain writable. */ 15466cdbceb1SKevin Wolf mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, 15476cdbceb1SKevin Wolf BDRV_O_RDWR, errp); 15484ef85a9cSKevin Wolf if (mirror_top_bs == NULL) { 1549cc19f177SVladimir Sementsov-Ogievskiy return NULL; 1550893f7ebaSPaolo Bonzini } 1551d3c8c674SKevin Wolf if (!filter_node_name) { 1552d3c8c674SKevin Wolf mirror_top_bs->implicit = true; 1553d3c8c674SKevin Wolf } 15544ef85a9cSKevin Wolf mirror_top_bs->total_sectors = bs->total_sectors; 1555228345bfSMax Reitz mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; 155680f5c33fSKevin Wolf mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | 155780f5c33fSKevin Wolf BDRV_REQ_NO_FALLBACK; 1558429076e8SMax Reitz bs_opaque = g_new0(MirrorBDSOpaque, 1); 1559429076e8SMax Reitz mirror_top_bs->opaque = bs_opaque; 1560893f7ebaSPaolo Bonzini 15614ef85a9cSKevin Wolf /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep 15627a25fcd0SMax Reitz * it alive until block_job_create() succeeds even if bs has no parent. */ 15634ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 15644ef85a9cSKevin Wolf bdrv_drained_begin(bs); 1565b2c2832cSKevin Wolf bdrv_append(mirror_top_bs, bs, &local_err); 15664ef85a9cSKevin Wolf bdrv_drained_end(bs); 15674ef85a9cSKevin Wolf 1568b2c2832cSKevin Wolf if (local_err) { 1569b2c2832cSKevin Wolf bdrv_unref(mirror_top_bs); 1570b2c2832cSKevin Wolf error_propagate(errp, local_err); 1571cc19f177SVladimir Sementsov-Ogievskiy return NULL; 1572b2c2832cSKevin Wolf } 1573b2c2832cSKevin Wolf 15744ef85a9cSKevin Wolf /* Make sure that the source is not resized while the job is running */ 157575859b94SJohn Snow s = block_job_create(job_id, driver, NULL, mirror_top_bs, 15764ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ, 15774ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 15784ef85a9cSKevin Wolf BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, 15794ef85a9cSKevin Wolf creation_flags, cb, opaque, errp); 15804ef85a9cSKevin Wolf if (!s) { 15814ef85a9cSKevin Wolf goto fail; 15824ef85a9cSKevin Wolf } 1583429076e8SMax Reitz bs_opaque->job = s; 1584429076e8SMax Reitz 15857a25fcd0SMax Reitz /* The block job now has a reference to this node */ 15867a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 15877a25fcd0SMax Reitz 15884ef85a9cSKevin Wolf s->mirror_top_bs = mirror_top_bs; 15894ef85a9cSKevin Wolf 15904ef85a9cSKevin Wolf /* No resize for the target either; while the mirror is still running, a 15914ef85a9cSKevin Wolf * consistent read isn't necessarily possible. We could possibly allow 15924ef85a9cSKevin Wolf * writes and graph modifications, though it would likely defeat the 15934ef85a9cSKevin Wolf * purpose of a mirror, so leave them blocked for now. 15944ef85a9cSKevin Wolf * 15954ef85a9cSKevin Wolf * In the case of active commit, things look a bit different, though, 15964ef85a9cSKevin Wolf * because the target is an already populated backing file in active use. 15974ef85a9cSKevin Wolf * We can allow anything except resize there.*/ 15984ef85a9cSKevin Wolf target_is_backing = bdrv_chain_contains(bs, target); 15994ef85a9cSKevin Wolf target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); 1600d861ab3aSKevin Wolf s->target = blk_new(s->common.job.aio_context, 1601d861ab3aSKevin Wolf BLK_PERM_WRITE | BLK_PERM_RESIZE | 16024ef85a9cSKevin Wolf (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), 16034ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | 16044ef85a9cSKevin Wolf (target_is_backing ? BLK_PERM_CONSISTENT_READ | 16054ef85a9cSKevin Wolf BLK_PERM_WRITE | 16064ef85a9cSKevin Wolf BLK_PERM_GRAPH_MOD : 0)); 1607d7086422SKevin Wolf ret = blk_insert_bs(s->target, target, errp); 1608d7086422SKevin Wolf if (ret < 0) { 16094ef85a9cSKevin Wolf goto fail; 1610d7086422SKevin Wolf } 1611045a2f82SFam Zheng if (is_mirror) { 1612045a2f82SFam Zheng /* XXX: Mirror target could be a NBD server of target QEMU in the case 1613045a2f82SFam Zheng * of non-shared block migration. To allow migration completion, we 1614045a2f82SFam Zheng * have to allow "inactivate" of the target BB. When that happens, we 1615045a2f82SFam Zheng * know the job is drained, and the vcpus are stopped, so no write 1616045a2f82SFam Zheng * operation will be performed. Block layer already has assertions to 1617045a2f82SFam Zheng * ensure that. */ 1618045a2f82SFam Zheng blk_set_force_allow_inactivate(s->target); 1619045a2f82SFam Zheng } 16209ff7f0dfSKevin Wolf blk_set_allow_aio_context_change(s->target, true); 1621e253f4b8SKevin Wolf 162209158f00SBenoît Canet s->replaces = g_strdup(replaces); 1623b952b558SPaolo Bonzini s->on_source_error = on_source_error; 1624b952b558SPaolo Bonzini s->on_target_error = on_target_error; 162503544a6eSFam Zheng s->is_none_mode = is_none_mode; 1626274fcceeSMax Reitz s->backing_mode = backing_mode; 1627481debaaSMax Reitz s->copy_mode = copy_mode; 16285bc361b8SFam Zheng s->base = base; 1629eee13dfeSPaolo Bonzini s->granularity = granularity; 163048ac0a4dSWen Congyang s->buf_size = ROUND_UP(buf_size, granularity); 16310fc9f8eaSFam Zheng s->unmap = unmap; 1632b49f7eadSWen Congyang if (auto_complete) { 1633b49f7eadSWen Congyang s->should_complete = true; 1634b49f7eadSWen Congyang } 1635b812f671SPaolo Bonzini 16360db6e54aSFam Zheng s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); 1637b8afb520SFam Zheng if (!s->dirty_bitmap) { 163888f9d1b3SKevin Wolf goto fail; 1639b8afb520SFam Zheng } 164010f3cd15SAlberto Garcia 164167b24427SAlberto Garcia ret = block_job_add_bdrv(&s->common, "source", bs, 0, 164267b24427SAlberto Garcia BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE | 164367b24427SAlberto Garcia BLK_PERM_CONSISTENT_READ, 164467b24427SAlberto Garcia errp); 164567b24427SAlberto Garcia if (ret < 0) { 164667b24427SAlberto Garcia goto fail; 164767b24427SAlberto Garcia } 164867b24427SAlberto Garcia 16494ef85a9cSKevin Wolf /* Required permissions are already taken with blk_new() */ 165076d554e2SKevin Wolf block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, 165176d554e2SKevin Wolf &error_abort); 165276d554e2SKevin Wolf 1653f3ede4b0SAlberto Garcia /* In commit_active_start() all intermediate nodes disappear, so 1654f3ede4b0SAlberto Garcia * any jobs in them must be blocked */ 16554ef85a9cSKevin Wolf if (target_is_backing) { 1656f3ede4b0SAlberto Garcia BlockDriverState *iter; 1657f3ede4b0SAlberto Garcia for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { 16584ef85a9cSKevin Wolf /* XXX BLK_PERM_WRITE needs to be allowed so we don't block 16594ef85a9cSKevin Wolf * ourselves at s->base (if writes are blocked for a node, they are 16604ef85a9cSKevin Wolf * also blocked for its backing file). The other options would be a 16614ef85a9cSKevin Wolf * second filter driver above s->base (== target). */ 16624ef85a9cSKevin Wolf ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, 16634ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, 16644ef85a9cSKevin Wolf errp); 16654ef85a9cSKevin Wolf if (ret < 0) { 16664ef85a9cSKevin Wolf goto fail; 16674ef85a9cSKevin Wolf } 1668f3ede4b0SAlberto Garcia } 1669ef53dc09SAlberto Garcia 1670ef53dc09SAlberto Garcia if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) { 1671ef53dc09SAlberto Garcia goto fail; 1672ef53dc09SAlberto Garcia } 1673f3ede4b0SAlberto Garcia } 167410f3cd15SAlberto Garcia 167512aa4082SMax Reitz QTAILQ_INIT(&s->ops_in_flight); 167612aa4082SMax Reitz 16775ccac6f1SJohn Snow trace_mirror_start(bs, s, opaque); 1678da01ff7fSKevin Wolf job_start(&s->common.job); 1679cc19f177SVladimir Sementsov-Ogievskiy 1680cc19f177SVladimir Sementsov-Ogievskiy return &s->common; 16814ef85a9cSKevin Wolf 16824ef85a9cSKevin Wolf fail: 16834ef85a9cSKevin Wolf if (s) { 16847a25fcd0SMax Reitz /* Make sure this BDS does not go away until we have completed the graph 16857a25fcd0SMax Reitz * changes below */ 16867a25fcd0SMax Reitz bdrv_ref(mirror_top_bs); 16877a25fcd0SMax Reitz 16884ef85a9cSKevin Wolf g_free(s->replaces); 16894ef85a9cSKevin Wolf blk_unref(s->target); 1690429076e8SMax Reitz bs_opaque->job = NULL; 1691e917e2cbSAlberto Garcia if (s->dirty_bitmap) { 1692e917e2cbSAlberto Garcia bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); 1693e917e2cbSAlberto Garcia } 16944ad35181SKevin Wolf job_early_fail(&s->common.job); 16954ef85a9cSKevin Wolf } 16964ef85a9cSKevin Wolf 1697f94dc3b4SMax Reitz bs_opaque->stop = true; 1698f94dc3b4SMax Reitz bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, 1699c1cef672SFam Zheng &error_abort); 17005fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 17017a25fcd0SMax Reitz 17027a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 1703cc19f177SVladimir Sementsov-Ogievskiy 1704cc19f177SVladimir Sementsov-Ogievskiy return NULL; 1705893f7ebaSPaolo Bonzini } 170603544a6eSFam Zheng 170771aa9867SAlberto Garcia void mirror_start(const char *job_id, BlockDriverState *bs, 170871aa9867SAlberto Garcia BlockDriverState *target, const char *replaces, 1709a1999b33SJohn Snow int creation_flags, int64_t speed, 1710a1999b33SJohn Snow uint32_t granularity, int64_t buf_size, 1711274fcceeSMax Reitz MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, 1712274fcceeSMax Reitz BlockdevOnError on_source_error, 171303544a6eSFam Zheng BlockdevOnError on_target_error, 1714481debaaSMax Reitz bool unmap, const char *filter_node_name, 1715481debaaSMax Reitz MirrorCopyMode copy_mode, Error **errp) 171603544a6eSFam Zheng { 171703544a6eSFam Zheng bool is_none_mode; 171803544a6eSFam Zheng BlockDriverState *base; 171903544a6eSFam Zheng 17204b80ab2bSJohn Snow if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { 17214b80ab2bSJohn Snow error_setg(errp, "Sync mode 'incremental' not supported"); 1722d58d8453SJohn Snow return; 1723d58d8453SJohn Snow } 172403544a6eSFam Zheng is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 1725760e0063SKevin Wolf base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; 1726a1999b33SJohn Snow mirror_start_job(job_id, bs, creation_flags, target, replaces, 1727274fcceeSMax Reitz speed, granularity, buf_size, backing_mode, 172851ccfa2dSFam Zheng on_source_error, on_target_error, unmap, NULL, NULL, 17296cdbceb1SKevin Wolf &mirror_job_driver, is_none_mode, base, false, 1730481debaaSMax Reitz filter_node_name, true, copy_mode, errp); 173103544a6eSFam Zheng } 173203544a6eSFam Zheng 1733cc19f177SVladimir Sementsov-Ogievskiy BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs, 173447970dfbSJohn Snow BlockDriverState *base, int creation_flags, 173547970dfbSJohn Snow int64_t speed, BlockdevOnError on_error, 17360db832f4SKevin Wolf const char *filter_node_name, 173778bbd910SFam Zheng BlockCompletionFunc *cb, void *opaque, 173878bbd910SFam Zheng bool auto_complete, Error **errp) 173903544a6eSFam Zheng { 17401ba79388SAlberto Garcia bool base_read_only; 1741cc67f4d1SJeff Cody Error *local_err = NULL; 1742cc19f177SVladimir Sementsov-Ogievskiy BlockJob *ret; 17434da83585SJeff Cody 17441ba79388SAlberto Garcia base_read_only = bdrv_is_read_only(base); 17454da83585SJeff Cody 17461ba79388SAlberto Garcia if (base_read_only) { 17471ba79388SAlberto Garcia if (bdrv_reopen_set_read_only(base, false, errp) < 0) { 1748cc19f177SVladimir Sementsov-Ogievskiy return NULL; 174920a63d2cSFam Zheng } 17501ba79388SAlberto Garcia } 17514da83585SJeff Cody 1752cc19f177SVladimir Sementsov-Ogievskiy ret = mirror_start_job( 1753cc19f177SVladimir Sementsov-Ogievskiy job_id, bs, creation_flags, base, NULL, speed, 0, 0, 175471aa9867SAlberto Garcia MIRROR_LEAVE_BACKING_CHAIN, 175551ccfa2dSFam Zheng on_error, on_error, true, cb, opaque, 17566cdbceb1SKevin Wolf &commit_active_job_driver, false, base, auto_complete, 1757481debaaSMax Reitz filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, 1758481debaaSMax Reitz &local_err); 17590fb6395cSMarkus Armbruster if (local_err) { 1760cc67f4d1SJeff Cody error_propagate(errp, local_err); 17614da83585SJeff Cody goto error_restore_flags; 17624da83585SJeff Cody } 17634da83585SJeff Cody 1764cc19f177SVladimir Sementsov-Ogievskiy return ret; 17654da83585SJeff Cody 17664da83585SJeff Cody error_restore_flags: 17674da83585SJeff Cody /* ignore error and errp for bdrv_reopen, because we want to propagate 17684da83585SJeff Cody * the original error */ 17691ba79388SAlberto Garcia if (base_read_only) { 17701ba79388SAlberto Garcia bdrv_reopen_set_read_only(base, true, NULL); 17711ba79388SAlberto Garcia } 1772cc19f177SVladimir Sementsov-Ogievskiy return NULL; 177303544a6eSFam Zheng } 1774