Reduce the number of I/O ops when doing COW
v3 cover Reduce the number of I/O ops when doing COW
Hi all,
here's a patch series that rewrites the copy-on-write code in the qcow2 driver to reduce the number of I/O operations.
This is version v3, please refer to the original e-mail for a complete description:
https://lists.gnu.org/archive/html/qemu-block/2017-05/msg00882.html
Regards,
Berto
v3:
- Patch 4: Over-allocate the read buffer in perform_cow() so both COW sectors are optimally aligned [Eric]
v2:
- Patch 1: Update commit message [Eric]
- Patch 7: Make sure that the number of iovs does not exceed IOV_MAX [Anton]
- Patch 7: Don't add zero-length buffers to the qiov in perform_cow()
v1: https://lists.gnu.org/archive/html/qemu-block/2017-05/msg00882.html
- Initial version
Output of git-backport-diff against v2:
Key:
[----] : patches are identical
[####] : number of functional differences between upstream/downstream patch [down] : patch is downstream-only
The flags [FC] indicate (F)unctional and (C)ontextual differences, respectively
001/7:[----] [--] 'qcow2: Remove unused Error variable in do_perform_cow()' 002/7:[----] [--] 'qcow2: Use unsigned int for both members of Qcow2COWRegion' 003/7:[----] [--] 'qcow2: Make perform_cow() call do_perform_cow() twice' 004/7:[0009] [FC] 'qcow2: Split do_perform_cow() into _read(), _encrypt() and _write()' 005/7:[0026] [FC] 'qcow2: Allow reading both COW regions with only one request' 006/7:[----] [-C] 'qcow2: Pass a QEMUIOVector to do_perform_cow_{read,write}()' 007/7:[----] [-C] 'qcow2: Merge the writing of the COW regions with the guest data'
Alberto Garcia (7):
qcow2: Remove unused Error variable in do_perform_cow()
qcow2: Use unsigned int for both members of Qcow2COWRegion
qcow2: Make perform_cow() call do_perform_cow() twice
qcow2: Split do_perform_cow() into _read(), _encrypt() and _write()
qcow2: Allow reading both COW regions with only one request
qcow2: Pass a QEMUIOVector to do_perform_cow_{read,write}()
qcow2: Merge the writing of the COW regions with the guest data
block/qcow2-cluster.c | 201 ++++++++++++++++++++++++++++++++++++++------------ block/qcow2.c | 64 +++++++++++++--- block/qcow2.h | 11 ++- 3 files changed, 216 insertions(+), 60 deletions(-)
Signature · 1 line
2.11.0
PATCH v3 1/7 qcow2: Remove unused Error variable in do_perform_cow()
We are using the return value of qcow2_encrypt_sectors() to detect problems but we are throwing away the returned Error since we have no way to report it to the user. Therefore we can simply get rid of the local Error variable and pass NULL instead.
Alternatively we could try to figure out a way to pass the original error instead of simply returning -EIO, but that would be more invasive, so let's keep the current approach.
--- block/qcow2-cluster.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
block/qcow2-cluster.c +1 -3
@@ -440,16 +440,14 @@ static int coroutine_fn do_perform_cow(BlockDriverState *bs, }if (bs->encrypted) { -Error *err =NULL ;int64_t sector = (src_cluster_offset + offset_in_cluster) >>BDRV_SECTOR_BITS ;assert (s->cipher);assert ((offset_in_cluster & ~BDRV_SECTOR_MASK ) ==0 );assert ((bytes & ~BDRV_SECTOR_MASK ) ==0 );if (qcow2_encrypt_sectors (s, sector, iov.iov_base, iov.iov_base, - bytes >>BDRV_SECTOR_BITS , true, &err) <0 ) { + bytes >>BDRV_SECTOR_BITS , true,NULL ) <0 ) { ret = -EIO ; -error_free (err); goto out; } }
Signature · 1 line
2.11.0
PATCH v3 2/7 qcow2: Use unsigned int for both members of Qcow2COWRegion
In reply to Alberto GarciaQcow2COWRegion has two attributes:
- The offset of the COW region from the start of the first cluster touched by the I/O request. Since it's always going to be positive and the maximum request size is at most INT_MAX, we can use a regular unsigned int to store this offset.
- The size of the COW region in bytes. This is guaranteed to be >= 0, so we should use an unsigned type instead.
In x86_64 this reduces the size of Qcow2COWRegion from 16 to 8 bytes. It will also help keep some assertions simpler now that we know that there are no negative numbers.
The prototype of do_perform_cow() is also updated to reflect these changes.
--- block/qcow2-cluster.c | 4 ++-- block/qcow2.h | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-)
block/qcow2-cluster.c +2 -2
@@ -406,8 +406,8 @@ int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num, static int coroutine_fn do_perform_cow(BlockDriverState *bs,uint64_t src_cluster_offset,uint64_t cluster_offset, -int offset_in_cluster, -int bytes) +unsigned offset_in_cluster, +unsigned bytes) {BDRVQcow2State *s = bs->opaque;QEMUIOVector qiov;
block/qcow2.h +2 -2
@@ -301,10 +301,10 @@ typedef struct Qcow2COWRegion {* Offset of the COW region in bytes from the start of the first cluster * touched by the request. */ -uint64_t offset; +unsigned offset;/** Number of bytes to copy */ -int nb_bytes; +unsigned nb_bytes; } Qcow2COWRegion;/**
Signature · 1 line
2.11.0
PATCH v3 3/7 qcow2: Make perform_cow() call do_perform_cow() twice
In reply to Alberto GarciaInstead of calling perform_cow() twice with a different COW region each time, call it just once and make perform_cow() handle both regions.
This patch simply moves code around. The next one will do the actual reordering of the COW operations.
--- block/qcow2-cluster.c | 38 +++++++++++++++++++++++--------------- 1 file changed, 23 insertions(+), 15 deletions(-)
block/qcow2-cluster.c +23 -15
@@ -414,6 +414,10 @@ static int coroutine_fn do_perform_cow(BlockDriverState *bs,struct iovec iov;int ret; +if (bytes ==0 ) { +return 0 ; + } + iov.iov_len = bytes; iov.iov_base =qemu_try_blockalign (bs, iov.iov_len);if (iov.iov_base ==NULL ) {
@@ -751,31 +755,40 @@ uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,return cluster_offset; } -static int perform_cow (BlockDriverState *bs,QCowL2Meta *m,Qcow2COWRegion *r) +static int perform_cow (BlockDriverState *bs,QCowL2Meta *m) {BDRVQcow2State *s = bs->opaque; +Qcow2COWRegion *start = &m->cow_start; +Qcow2COWRegion *end = &m->cow_end;int ret; -if (r->nb_bytes ==0 ) { +if (start->nb_bytes ==0 && end->nb_bytes ==0 ) {return 0 ; }qemu_co_mutex_unlock (&s->lock); - ret =do_perform_cow (bs, m->offset, m->alloc_offset, r->offset, r->nb_bytes); + ret =do_perform_cow (bs, m->offset, m->alloc_offset, + start->offset, start->nb_bytes); +if (ret <0 ) { + goto fail; + } + + ret =do_perform_cow (bs, m->offset, m->alloc_offset, + end->offset, end->nb_bytes); + +fail:qemu_co_mutex_lock (&s->lock); -if (ret <0 ) { -return ret; - } -/* * Before we update the L2 table to actually point to the new cluster, we * need to be sure that the refcounts have been increased and COW was * handled. */ -qcow2_cache_depends_on_flush (s->l2_table_cache); +if (ret ==0 ) { +qcow2_cache_depends_on_flush (s->l2_table_cache); + } -return 0 ; +return ret; }int qcow2_alloc_cluster_link_l2 (BlockDriverState *bs,QCowL2Meta *m)
@@ -795,12 +808,7 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) }/* copy content of unmodified sectors */ - ret =perform_cow (bs, m, &m->cow_start); -if (ret <0 ) { - goto err; - } - - ret =perform_cow (bs, m, &m->cow_end); + ret =perform_cow (bs, m);if (ret <0 ) { goto err; }
Signature · 1 line
2.11.0
PATCH v3 4/7 qcow2: Split do_perform_cow() into _read(), _encrypt() and _write()
In reply to Alberto GarciaThis patch splits do_perform_cow() into three separate functions to read, encrypt and write the COW regions.
perform_cow() can now read both regions first, then encrypt them and finally write them to disk. The memory allocation is also done in this function now, using one single buffer large enough to hold both regions.
--- block/qcow2-cluster.c | 117 +++++++++++++++++++++++++++++++++++++------------- 1 file changed, 87 insertions(+), 30 deletions(-)
block/qcow2-cluster.c +87 -30
@@ -403,34 +403,26 @@ int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,return 0 ; } -static int coroutine_fn do_perform_cow(BlockDriverState *bs, -uint64_t src_cluster_offset, -uint64_t cluster_offset, -unsigned offset_in_cluster, -unsigned bytes) +static int coroutine_fn do_perform_cow_read(BlockDriverState *bs, +uint64_t src_cluster_offset, +unsigned offset_in_cluster, +uint8_t *buffer, +unsigned bytes) { -BDRVQcow2State *s = bs->opaque;QEMUIOVector qiov; -struct iovec iov; +struct iovec iov = { .iov_base = buffer, .iov_len = bytes };int ret;if (bytes ==0 ) {return 0 ; } - iov.iov_len = bytes; - iov.iov_base =qemu_try_blockalign (bs, iov.iov_len); -if (iov.iov_base ==NULL ) { -return -ENOMEM ; - } -qemu_iovec_init_external (&qiov, &iov,1 );BLKDBG_EVENT (bs->file,BLKDBG_COW_READ );if (!bs->drv) { - ret = -ENOMEDIUM ; - goto out; +return -ENOMEDIUM ; }/* Call .bdrv_co_readv() directly instead of using the public block-layer
@@ -440,39 +432,63 @@ static int coroutine_fn do_perform_cow(BlockDriverState *bs, ret = bs->drv->bdrv_co_preadv (bs, src_cluster_offset + offset_in_cluster, bytes, &qiov,0 );if (ret <0 ) { - goto out; +return ret; } -if (bs->encrypted) { +return 0 ; +} + +static bool coroutine_fn do_perform_cow_encrypt(BlockDriverState *bs, +uint64_t src_cluster_offset, +unsigned offset_in_cluster, +uint8_t *buffer, +unsigned bytes) +{ +if (bytes && bs->encrypted) { +BDRVQcow2State *s = bs->opaque;int64_t sector = (src_cluster_offset + offset_in_cluster) >>BDRV_SECTOR_BITS ;assert (s->cipher);assert ((offset_in_cluster & ~BDRV_SECTOR_MASK ) ==0 );assert ((bytes & ~BDRV_SECTOR_MASK ) ==0 ); -if (qcow2_encrypt_sectors (s, sector, iov.iov_base, iov.iov_base, +if (qcow2_encrypt_sectors (s, sector, buffer, buffer, bytes >>BDRV_SECTOR_BITS , true,NULL ) <0 ) { - ret = -EIO ; - goto out; +return false; } } +return true; +} + +static int coroutine_fn do_perform_cow_write(BlockDriverState *bs, +uint64_t cluster_offset, +unsigned offset_in_cluster, +uint8_t *buffer, +unsigned bytes) +{ +QEMUIOVector qiov; +struct iovec iov = { .iov_base = buffer, .iov_len = bytes }; +int ret; + +if (bytes ==0 ) { +return 0 ; + } + +qemu_iovec_init_external (&qiov, &iov,1 ); ret =qcow2_pre_write_overlap_check (bs,0 , cluster_offset + offset_in_cluster, bytes);if (ret <0 ) { - goto out; +return ret; }BLKDBG_EVENT (bs->file,BLKDBG_COW_WRITE ); ret =bdrv_co_pwritev (bs->file, cluster_offset + offset_in_cluster, bytes, &qiov,0 );if (ret <0 ) { - goto out; +return ret; } - ret =0 ; -out: -qemu_vfree (iov.iov_base); -return ret; +return 0 ; }
@@ -760,22 +776,62 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)BDRVQcow2State *s = bs->opaque;Qcow2COWRegion *start = &m->cow_start;Qcow2COWRegion *end = &m->cow_end; +unsigned buffer_size; +uint8_t *start_buffer, *end_buffer;int ret; +assert (start->nb_bytes <=UINT_MAX - end->nb_bytes); +if (start->nb_bytes ==0 && end->nb_bytes ==0 ) {return 0 ; } +/* Reserve a buffer large enough to store the data from both the +* start and end COW regions. Add some padding in the middle if +* necessary to make sure that the end region is optimally aligned */ + buffer_size =QEMU_ALIGN_UP (start->nb_bytes,bdrv_opt_mem_align (bs)) + + end->nb_bytes; + start_buffer =qemu_try_blockalign (bs, buffer_size); +if (start_buffer ==NULL ) { +return -ENOMEM ; + } +/* The part of the buffer where the end region is located */ + end_buffer = start_buffer + buffer_size - end->nb_bytes; +qemu_co_mutex_unlock (&s->lock); - ret =do_perform_cow (bs, m->offset, m->alloc_offset, - start->offset, start->nb_bytes); +/* First we read the existing data from both COW regions */ + ret =do_perform_cow_read (bs, m->offset, start->offset, + start_buffer, start->nb_bytes);if (ret <0 ) { goto fail; } - ret =do_perform_cow (bs, m->offset, m->alloc_offset, - end->offset, end->nb_bytes); + ret =do_perform_cow_read (bs, m->offset, end->offset, + end_buffer, end->nb_bytes); +if (ret <0 ) { + goto fail; + } + +/* Encrypt the data if necessary before writing it */ +if (bs->encrypted) { +if (!do_perform_cow_encrypt (bs, m->offset, start->offset, + start_buffer, start->nb_bytes) || + !do_perform_cow_encrypt (bs, m->offset, end->offset, + end_buffer, end->nb_bytes)) { + ret = -EIO ; + goto fail; + } + } + +/* And now we can write everything */ + ret =do_perform_cow_write (bs, m->alloc_offset, start->offset, + start_buffer, start->nb_bytes); +if (ret <0 ) { + goto fail; + } + ret =do_perform_cow_write (bs, m->alloc_offset, end->offset, + end_buffer, end->nb_bytes); fail:qemu_co_mutex_lock (&s->lock);
@@ -788,6 +844,7 @@ fail:qcow2_cache_depends_on_flush (s->l2_table_cache); } +qemu_vfree (start_buffer);return ret; }
Signature · 1 line
2.11.0
PATCH v3 5/7 qcow2: Allow reading both COW regions with only one request
In reply to Alberto GarciaReading both COW regions requires two separate requests, but it's perfectly possible to merge them and perform only one. This generally improves performance, particularly on rotating disk drives. The downside is that the data in the middle region is read but discarded.
This patch takes a conservative approach and only merges reads when the size of the middle region is <= 16KB.
--- block/qcow2-cluster.c | 51 ++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 38 insertions(+), 13 deletions(-)
block/qcow2-cluster.c +38 -13
@@ -777,20 +777,38 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)Qcow2COWRegion *start = &m->cow_start;Qcow2COWRegion *end = &m->cow_end;unsigned buffer_size; +unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); +bool merge_reads;uint8_t *start_buffer, *end_buffer;int ret;assert (start->nb_bytes <=UINT_MAX - end->nb_bytes); +assert (start->nb_bytes + end->nb_bytes <=UINT_MAX - data_bytes); +assert (start->offset + start->nb_bytes <= end->offset);if (start->nb_bytes ==0 && end->nb_bytes ==0 ) {return 0 ; } -/* Reserve a buffer large enough to store the data from both the -* start and end COW regions. Add some padding in the middle if -* necessary to make sure that the end region is optimally aligned */ - buffer_size =QEMU_ALIGN_UP (start->nb_bytes,bdrv_opt_mem_align (bs)) + - end->nb_bytes; +/* If we have to read both the start and end COW regions and the +* middle region is not too large then perform just one read +* operation */ + merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <=16384 ; +if (merge_reads) { + buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; + }else { +/* If we have to do two reads, add some padding in the middle +* if necessary to make sure that the end region is optimally +* aligned. */ +size_t align =bdrv_opt_mem_align (bs); +assert (align >0 && align <=UINT_MAX ); +assert (QEMU_ALIGN_UP (start->nb_bytes, align) <= +UINT_MAX - end->nb_bytes); + buffer_size =QEMU_ALIGN_UP (start->nb_bytes, align) + end->nb_bytes; + } + +/* Reserve a buffer large enough to store all the data that we're +* going to read */ start_buffer =qemu_try_blockalign (bs, buffer_size);if (start_buffer ==NULL ) {return -ENOMEM ;
@@ -799,15 +817,22 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) end_buffer = start_buffer + buffer_size - end->nb_bytes;qemu_co_mutex_unlock (&s->lock); -/* First we read the existing data from both COW regions */ - ret =do_perform_cow_read (bs, m->offset, start->offset, - start_buffer, start->nb_bytes); -if (ret <0 ) { - goto fail; - } +/* First we read the existing data from both COW regions. We +* either read the whole region in one go, or the start and end +* regions separately. */ +if (merge_reads) { + ret =do_perform_cow_read (bs, m->offset, start->offset, + start_buffer, buffer_size); + }else { + ret =do_perform_cow_read (bs, m->offset, start->offset, + start_buffer, start->nb_bytes); +if (ret <0 ) { + goto fail; + } - ret =do_perform_cow_read (bs, m->offset, end->offset, - end_buffer, end->nb_bytes); + ret =do_perform_cow_read (bs, m->offset, end->offset, + end_buffer, end->nb_bytes); + }if (ret <0 ) { goto fail; }
Signature · 1 line
2.11.0
PATCH v3 6/7 qcow2: Pass a QEMUIOVector to do_perform_cow_{read, write}()
In reply to Alberto GarciaInstead of passing a single buffer pointer to do_perform_cow_write(), pass a QEMUIOVector. This will allow us to merge the write requests for the COW regions and the actual data into a single one.
Although do_perform_cow_read() does not strictly need to change its API, we're doing it here as well for consistency.
--- block/qcow2-cluster.c | 51 ++++++++++++++++++++++++--------------------------- 1 file changed, 24 insertions(+), 27 deletions(-)
block/qcow2-cluster.c +24 -27
@@ -406,19 +406,14 @@ int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num, static int coroutine_fn do_perform_cow_read(BlockDriverState *bs,uint64_t src_cluster_offset,unsigned offset_in_cluster, -uint8_t *buffer, -unsigned bytes) +QEMUIOVector *qiov) { -QEMUIOVector qiov; -struct iovec iov = { .iov_base = buffer, .iov_len = bytes };int ret; -if (bytes ==0 ) { +if (qiov->size ==0 ) {return 0 ; } -qemu_iovec_init_external (&qiov, &iov,1 ); -BLKDBG_EVENT (bs->file,BLKDBG_COW_READ );if (!bs->drv) {
@@ -430,7 +425,7 @@ static int coroutine_fn do_perform_cow_read(BlockDriverState *bs,* which can lead to deadlock when block layer copy-on-read is enabled. */ ret = bs->drv->bdrv_co_preadv (bs, src_cluster_offset + offset_in_cluster, - bytes, &qiov,0 ); + qiov->size, qiov,0 );if (ret <0 ) {return ret; }
@@ -462,28 +457,23 @@ static bool coroutine_fn do_perform_cow_encrypt(BlockDriverState *bs, static int coroutine_fn do_perform_cow_write(BlockDriverState *bs,uint64_t cluster_offset,unsigned offset_in_cluster, -uint8_t *buffer, -unsigned bytes) +QEMUIOVector *qiov) { -QEMUIOVector qiov; -struct iovec iov = { .iov_base = buffer, .iov_len = bytes };int ret; -if (bytes ==0 ) { +if (qiov->size ==0 ) {return 0 ; } -qemu_iovec_init_external (&qiov, &iov,1 ); - ret =qcow2_pre_write_overlap_check (bs,0 , - cluster_offset + offset_in_cluster, bytes); + cluster_offset + offset_in_cluster, qiov->size);if (ret <0 ) {return ret; }BLKDBG_EVENT (bs->file,BLKDBG_COW_WRITE ); ret =bdrv_co_pwritev (bs->file, cluster_offset + offset_in_cluster, - bytes, &qiov,0 ); + qiov->size, qiov,0 );if (ret <0 ) {return ret; }
@@ -780,6 +770,7 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)unsigned data_bytes = end->offset - (start->offset + start->nb_bytes);bool merge_reads;uint8_t *start_buffer, *end_buffer; +QEMUIOVector qiov;int ret;assert (start->nb_bytes <=UINT_MAX - end->nb_bytes);
@@ -816,22 +807,25 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)/* The part of the buffer where the end region is located */ end_buffer = start_buffer + buffer_size - end->nb_bytes; +qemu_iovec_init (&qiov,3 ); +qemu_co_mutex_unlock (&s->lock);/* First we read the existing data from both COW regions. We * either read the whole region in one go, or the start and end * regions separately. */ if (merge_reads) { - ret =do_perform_cow_read (bs, m->offset, start->offset, - start_buffer, buffer_size); +qemu_iovec_add (&qiov, start_buffer, buffer_size); + ret =do_perform_cow_read (bs, m->offset, start->offset, &qiov); }else { - ret =do_perform_cow_read (bs, m->offset, start->offset, - start_buffer, start->nb_bytes); +qemu_iovec_add (&qiov, start_buffer, start->nb_bytes); + ret =do_perform_cow_read (bs, m->offset, start->offset, &qiov);if (ret <0 ) { goto fail; } - ret =do_perform_cow_read (bs, m->offset, end->offset, - end_buffer, end->nb_bytes); +qemu_iovec_reset (&qiov); +qemu_iovec_add (&qiov, end_buffer, end->nb_bytes); + ret =do_perform_cow_read (bs, m->offset, end->offset, &qiov); }if (ret <0 ) { goto fail;
@@ -849,14 +843,16 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) }/* And now we can write everything */ - ret =do_perform_cow_write (bs, m->alloc_offset, start->offset, - start_buffer, start->nb_bytes); +qemu_iovec_reset (&qiov); +qemu_iovec_add (&qiov, start_buffer, start->nb_bytes); + ret =do_perform_cow_write (bs, m->alloc_offset, start->offset, &qiov);if (ret <0 ) { goto fail; } - ret =do_perform_cow_write (bs, m->alloc_offset, end->offset, - end_buffer, end->nb_bytes); +qemu_iovec_reset (&qiov); +qemu_iovec_add (&qiov, end_buffer, end->nb_bytes); + ret =do_perform_cow_write (bs, m->alloc_offset, end->offset, &qiov); fail:qemu_co_mutex_lock (&s->lock);
@@ -870,6 +866,7 @@ fail: }qemu_vfree (start_buffer); +qemu_iovec_destroy (&qiov);return ret; }
Signature · 1 line
2.11.0
PATCH v3 7/7 qcow2: Merge the writing of the COW regions with the guest data
In reply to Alberto GarciaIf the guest tries to write data that results on the allocation of a new cluster, instead of writing the guest data first and then the data from the COW regions, write everything together using one single I/O operation.
This can improve the write performance by 25% or more, depending on several factors such as the media type, the cluster size and the I/O request size.
--- block/qcow2-cluster.c | 38 ++++++++++++++++++++++-------- block/qcow2.c | 64 +++++++++++++++++++++++++++++++++++++++++++-------- block/qcow2.h | 7 ++++++ 3 files changed, 90 insertions(+), 19 deletions(-)
block/qcow2-cluster.c +29 -9
@@ -776,6 +776,7 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)assert (start->nb_bytes <=UINT_MAX - end->nb_bytes);assert (start->nb_bytes + end->nb_bytes <=UINT_MAX - data_bytes);assert (start->offset + start->nb_bytes <= end->offset); +assert (!m->data_qiov || m->data_qiov->size == data_bytes);if (start->nb_bytes ==0 && end->nb_bytes ==0 ) {return 0 ;
@@ -842,17 +843,36 @@ static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) } } -/* And now we can write everything */ -qemu_iovec_reset (&qiov); -qemu_iovec_add (&qiov, start_buffer, start->nb_bytes); - ret =do_perform_cow_write (bs, m->alloc_offset, start->offset, &qiov); -if (ret <0 ) { - goto fail; +/* And now we can write everything. If we have the guest data we +* can write everything in one single operation */ +if (m->data_qiov) { +qemu_iovec_reset (&qiov); +if (start->nb_bytes) { +qemu_iovec_add (&qiov, start_buffer, start->nb_bytes); + } +qemu_iovec_concat (&qiov, m->data_qiov,0 , data_bytes); +if (end->nb_bytes) { +qemu_iovec_add (&qiov, end_buffer, end->nb_bytes); + } +/* NOTE: we have a write_aio blkdebug event here followed by +* a cow_write one in do_perform_cow_write(), but there's only +* one single I/O operation */ +BLKDBG_EVENT (bs->file,BLKDBG_WRITE_AIO ); + ret =do_perform_cow_write (bs, m->alloc_offset, start->offset, &qiov); + }else { +/* If there's no guest data then write both COW regions separately */ +qemu_iovec_reset (&qiov); +qemu_iovec_add (&qiov, start_buffer, start->nb_bytes); + ret =do_perform_cow_write (bs, m->alloc_offset, start->offset, &qiov); +if (ret <0 ) { + goto fail; + } + +qemu_iovec_reset (&qiov); +qemu_iovec_add (&qiov, end_buffer, end->nb_bytes); + ret =do_perform_cow_write (bs, m->alloc_offset, end->offset, &qiov); } -qemu_iovec_reset (&qiov); -qemu_iovec_add (&qiov, end_buffer, end->nb_bytes); - ret =do_perform_cow_write (bs, m->alloc_offset, end->offset, &qiov); fail:qemu_co_mutex_lock (&s->lock);
block/qcow2.c +54 -10
76 lines not on this page · raw message
block/qcow2.h +7 -0
@@ -343,6 +343,13 @@ typedef struct QCowL2Meta*/ Qcow2COWRegion cow_end; +/** +* The I/O vector with the data from the actual guest write request. +* If non-NULL, this is meant to be merged together with the data +* from @cow_start and @cow_end into one single write operation. +*/ +QEMUIOVector *data_qiov; +/** Pointer to next L2Meta of the same write request */ struct QCowL2Meta *next;
Signature · 1 line
2.11.0