make bdrv_get_block_status byte-based
v6 cover make bdrv_get_block_status byte-based
There are patches floating around to add NBD_CMD_BLOCK_STATUS, but NBD wants to report status on byte granularity (even if the reporting will probably be naturally aligned to sectors or even much higher levels). I've therefore started the task of converting our block status code to report at a byte granularity rather than sectors.
Now that 2.11 is open, I'm rebasing/reposting the remaining patches.
The overall conversion currently looks like:
part 1: bdrv_is_allocated (merged, commit 51b0a488)
part 2: dirty-bitmap (merged, commit ca759622)
part 3: bdrv_get_block_status (this series, v5 at [1])
part 4: .bdrv_co_block_status (v3 is posted [2], mostly reviewed, but needs rebase)
Available as a tag at:
git fetch git://repo.or.cz/qemu/ericb.git nbd-byte-status-v6
Based-on: <[email protected]> ([PATCH 0/3] qcow2: Fix preallocated truncation)
(or more precisely, based on Max' block branch that includes that and other iotest patches, as I was hitting iotest failures without that branch; however, I suspect that the dependency is weak enough that this will apply just fine on master at the moment)
[1] https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg00535.html
[2] https://lists.gnu.org/archive/html/qemu-devel/2017-09/msg03812.html
Since v5:
- rebase to master, since copy-on-read patches (commit 46174339) that landed there conflict with v5
- tweak qcow2 is_zero() to not round unnecessarily [Kevin]
- major interface overhaul: instead of returning a sector-based offset in a 64-bit return type (which then requires post-call munging to get to byte offsets), return a byte-based offset via a parameter. Lots of churn in the interdiff, but the end result looks better [Kevin]
- other minor fixes pointed out during review, such as -O2 compilation, improved comments, ...
001/24:[0031] [FC] 'block: Allow NULL file for bdrv_get_block_status()' 002/24:[0029] [FC] 'block: Add flag to avoid wasted work in bdrv_is_allocated()' 003/24:[----] [--] 'block: Make bdrv_round_to_clusters() signature more useful' 004/24:[0005] [FC] 'qcow2: Switch is_zero_sectors() to byte-based' 005/24:[----] [--] 'block: Switch bdrv_make_zero() to byte-based' 006/24:[----] [--] 'qemu-img: Switch get_block_status() to byte-based' 007/24:[0048] [FC] 'block: Convert bdrv_get_block_status() to bytes' 008/24:[0105] [FC] 'block: Switch bdrv_co_get_block_status() to byte-based' 009/24:[0037] [FC] 'block: Switch BdrvCoGetBlockStatusData to byte-based' 010/24:[0036] [FC] 'block: Switch bdrv_common_block_status_above() to byte-based' 011/24:[0056] [FC] 'block: Switch bdrv_co_get_block_status_above() to byte-based' 012/24:[0066] [FC] 'block: Convert bdrv_get_block_status_above() to bytes' 013/24:[----] [-C] 'qemu-img: Simplify logic in img_compare()' 014/24:[0002] [FC] 'qemu-img: Speed up compare on pre-allocated larger file' 015/24:[----] [--] 'qemu-img: Add find_nonzero()'
016/24:[----] [--] 'qemu-img: Drop redundant error message in compare' 017/24:[----] [--] 'qemu-img: Change check_empty_sectors() to byte-based' 018/24:[----] [--] 'qemu-img: Change compare_sectors() to be byte-based' 019/24:[----] [--] 'qemu-img: Change img_rebase() to be byte-based' 020/24:[0014] [FC] 'qemu-img: Change img_compare() to be byte-based' 021/24:[0055] [FC] 'block: Align block status requests' 022/24:[----] [--] 'block: Relax bdrv_aligned_preadv() assertion' 023/24:[down] 'qcow2: Relax is_zero() assertion'
024/24:[----] [--] 'qemu-io: Relax 'alloc' now that block-status doesn't assert'
Eric Blake (24):
block: Allow NULL file for bdrv_get_block_status()
block: Add flag to avoid wasted work in bdrv_is_allocated()
block: Make bdrv_round_to_clusters() signature more useful
qcow2: Switch is_zero_sectors() to byte-based
block: Switch bdrv_make_zero() to byte-based
qemu-img: Switch get_block_status() to byte-based
block: Convert bdrv_get_block_status() to bytes
block: Switch bdrv_co_get_block_status() to byte-based
block: Switch BdrvCoGetBlockStatusData to byte-based
block: Switch bdrv_common_block_status_above() to byte-based
block: Switch bdrv_co_get_block_status_above() to byte-based
block: Convert bdrv_get_block_status_above() to bytes
qemu-img: Simplify logic in img_compare()
qemu-img: Speed up compare on pre-allocated larger file
qemu-img: Add find_nonzero()
qemu-img: Drop redundant error message in compare
qemu-img: Change check_empty_sectors() to byte-based
qemu-img: Change compare_sectors() to be byte-based
qemu-img: Change img_rebase() to be byte-based
qemu-img: Change img_compare() to be byte-based
block: Align block status requests
block: Relax bdrv_aligned_preadv() assertion
qcow2: Relax is_zero() assertion
qemu-io: Relax 'alloc' now that block-status doesn't assert
include/block/block.h | 29 ++-- include/block/block_int.h | 11 +- block/io.c | 332 +++++++++++++++++++++++---------------- block/blkdebug.c | 13 +- block/mirror.c | 26 ++-- block/qcow2-cluster.c | 2 +- block/qcow2.c | 53 +++---- qemu-img.c | 381 ++++++++++++++++++++------------------------- qemu-io-cmds.c | 13 -- block/trace-events | 2 +- tests/qemu-iotests/074.out | 2 - tests/qemu-iotests/177 | 12 +- tests/qemu-iotests/177.out | 19 ++- 13 files changed, 459 insertions(+), 436 deletions(-)
Signature · 1 line
2.13.6
PATCH v6 1/24 block: Allow NULL file for bdrv_get_block_status()
Not all callers care about which BDS owns the mapping for a given range of the file. This patch merely simplifies the callers by consolidating the logic in the common call point, while guaranteeing a non-NULL file to all the driver callbacks, for no semantic change. The only caller that does not care about pnum is bdrv_is_allocated, as invoked by vvfat; we can likewise add assertions that the rest of the stack does not have to worry about a NULL pnum.
Furthermore, this will also set the stage for a future cleanup: when a caller does not care about which BDS owns an offset, it would be nice to allow the driver to optimize things to not have to return BDRV_BLOCK_OFFSET_VALID in the first place. In the case of fragmented allocation (for example, it's fairly easy to create a qcow2 image where consecutive guest addresses are not at consecutive host addresses), the current contract requires bdrv_get_block_status() to clamp *pnum to the limit where host addresses are no longer consecutive, but allowing a NULL file means that *pnum could be set to the full length of known-allocated data.
---
v6: drop local_pnum experiment, tweak label indent [Kevin]
v5: use second label for cleaner exit logic [John], use local_pnum
v4: only context changes
v3: rebase to recent changes (qcow2_measure), dropped R-b
v2: use local variable and final transfer, rather than assignment
of parameter to local
[previously in different series]:
v2: new patch, https://lists.gnu.org/archive/html/qemu-devel/2017-05/msg05645.html
--- include/block/block_int.h | 10 ++++++---- block/io.c | 49 ++++++++++++++++++++++++++--------------------- block/mirror.c | 3 +-- block/qcow2.c | 8 ++------ qemu-img.c | 10 ++++------ 5 files changed, 40 insertions(+), 40 deletions(-)
include/block/block_int.h +6 -4
@@ -202,10 +202,12 @@ struct BlockDriver { int64_t offset, int bytes);/* -* Building block for bdrv_block_status[_above]. The driver should -* answer only according to the current layer, and should not -* set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW. See block.h -* for the meaning of _DATA, _ZERO, and _OFFSET_VALID. +* Building block for bdrv_block_status[_above] and +* bdrv_is_allocated[_above]. The driver should answer only +* according to the current layer, and should not set +* BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW. See block.h +* for the meaning of _DATA, _ZERO, and _OFFSET_VALID. The block +* layer guarantees non-NULL pnum and file. */ int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
block/io.c +27 -22
126 lines not on this page · raw message
block/mirror.c +1 -2
@@ -390,7 +390,6 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)int io_sectors;unsigned int io_bytes;int64_t io_bytes_acct; -BlockDriverState *file;enum MirrorMethod {MIRROR_METHOD_COPY ,MIRROR_METHOD_ZERO ,
@@ -401,7 +400,7 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) ret =bdrv_get_block_status_above (source,NULL , offset >>BDRV_SECTOR_BITS , nb_chunks * sectors_per_chunk, - &io_sectors, &file); + &io_sectors,NULL ); io_bytes = io_sectors *BDRV_SECTOR_SIZE ;if (ret <0 ) { io_bytes =MIN (nb_chunks * s->granularity, max_io_bytes);
block/qcow2.c +2 -6
@@ -3003,7 +3003,6 @@ static bool is_zero_sectors(BlockDriverState *bs, int64_t start, uint32_t count) {int nr; -BlockDriverState *file;int64_t res;if (start + count > bs->total_sectors) {
@@ -3013,8 +3012,7 @@ static bool is_zero_sectors(BlockDriverState *bs, int64_t start,if (!count) {return true; } - res =bdrv_get_block_status_above (bs,NULL , start, count, - &nr, &file); + res =bdrv_get_block_status_above (bs,NULL , start, count, &nr,NULL );return res >=0 && (res &BDRV_BLOCK_ZERO ) && nr == count; }
@@ -3732,13 +3730,11 @@ static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, offset += pnum * BDRV_SECTOR_SIZE) {int nb_sectors =MIN (ssize - offset,BDRV_REQUEST_MAX_BYTES ) /BDRV_SECTOR_SIZE ; -BlockDriverState *file;int64_t ret; ret =bdrv_get_block_status_above (in_bs,NULL , offset >>BDRV_SECTOR_BITS , - nb_sectors, - &pnum, &file); + nb_sectors, &pnum,NULL );if (ret <0 ) {error_setg_errno (&local_err, -ret,"Unable to get block status" );
qemu-img.c +4 -6
@@ -1375,7 +1375,6 @@ static int img_compare(int argc, char **argv)for (;;) {int64_t status1, status2; -BlockDriverState *file; nb_sectors =sectors_to_process (total_sectors, sector_num);if (nb_sectors <=0 ) {
@@ -1383,7 +1382,7 @@ static int img_compare(int argc, char **argv) } status1 =bdrv_get_block_status_above (bs1,NULL , sector_num, total_sectors1 - sector_num, - &pnum1, &file); + &pnum1,NULL );if (status1 <0 ) { ret =3 ;error_report ("Sector allocation test failed for %s" , filename1);
@@ -1393,7 +1392,7 @@ static int img_compare(int argc, char **argv) status2 =bdrv_get_block_status_above (bs2,NULL , sector_num, total_sectors2 - sector_num, - &pnum2, &file); + &pnum2,NULL );if (status2 <0 ) { ret =3 ;error_report ("Sector allocation test failed for %s" , filename2);
@@ -1599,15 +1598,14 @@ static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num) n =MIN (s->total_sectors - sector_num,BDRV_REQUEST_MAX_SECTORS );if (s->sector_next_status <= sector_num) { -BlockDriverState *file;if (s->target_has_backing) { ret =bdrv_get_block_status (blk_bs (s->src[src_cur]), sector_num - src_cur_offset, - n, &n, &file); + n, &n,NULL ); }else { ret =bdrv_get_block_status_above (blk_bs (s->src[src_cur]),NULL , sector_num - src_cur_offset, - n, &n, &file); + n, &n,NULL ); }if (ret <0 ) {return ret;
Signature · 1 line
2.13.6
PATCH v6 2/24 block: Add flag to avoid wasted work in bdrv_is_allocated()
In reply to Eric BlakeNot all callers care about which BDS owns the mapping for a given range of the file, or where the zeroes lie within that mapping. In particular, bdrv_is_allocated() cares more about finding the largest run of allocated data from the guest perspective, whether or not that data is consecutive from the host perspective, and whether or not the data reads as zero. Therefore, doing subsequent refinements such as checking how much of the format-layer allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is wasted work - in the best case, it just costs extra CPU cycles during a single bdrv_is_allocated(), but in the worst case, it results in a smaller *pnum, and forces callers to iterate through more status probes when visiting the entire file for even more extra CPU cycles.
This patch only optimizes the block layer (no behavior change when want_zero is true, but skip unnecessary effort when it is false). Then when subsequent patches tweak the driver callback to be byte-based, we can also pass this hint through to the driver.
Tweak BdrvCoGetBlockStatusData to declare arguments in parameter order, rather than mixing things up (minimizing padding is not necessary here).
---
v6: rebase to previous changes, s/mapping/need_zero/, tweak struct layout, drop R-b
v5: tweak commit message and one comment, rebase to previous changes, minor enough to still add R-b
v4: only context changes
v3: s/allocation/mapping/ and flip sense of bool
v2: new patch
--- block/io.c | 57 +++++++++++++++++++++++++++++++++++++++++---------------- 1 file changed, 41 insertions(+), 16 deletions(-)
block/io.c +41 -16
@@ -1749,10 +1749,11 @@ int bdrv_flush_all(void)typedef struct BdrvCoGetBlockStatusData {BlockDriverState *bs;BlockDriverState *base; -BlockDriverState **file; +bool want_zero;int64_t sector_num;int nb_sectors;int *pnum; +BlockDriverState **file;int64_t ret;bool done; }BdrvCoGetBlockStatusData ;
@@ -1788,6 +1789,11 @@ int64_t coroutine_fn bdrv_co_get_block_status_from_backing(BlockDriverState *bs,* Drivers not implementing the functionality are assumed to not support * backing files, hence all their sectors are reported as allocated. * +* If 'want_zero' is true, the caller is querying for mapping purposes, +* and the result should include BDRV_BLOCK_OFFSET_VALID and +* BDRV_BLOCK_ZERO where possible; otherwise, the result may omit those +* bits particularly if it allows for a larger value in 'pnum'. +* * If 'sector_num' is beyond the end of the disk image the return value is * BDRV_BLOCK_EOF and 'pnum' is set to 0. *
@@ -1804,6 +1810,7 @@ int64_t coroutine_fn bdrv_co_get_block_status_from_backing(BlockDriverState *bs,* is allocated in. */ static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, +bool want_zero,int64_t sector_num,int nb_sectors,int *pnum,BlockDriverState **file)
@@ -1858,31 +1865,34 @@ static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,if (ret &BDRV_BLOCK_RAW ) {assert (ret &BDRV_BLOCK_OFFSET_VALID && local_file); - ret =bdrv_co_get_block_status (local_file, ret >>BDRV_SECTOR_BITS , + ret =bdrv_co_get_block_status (local_file, want_zero, + ret >>BDRV_SECTOR_BITS , *pnum, pnum, &local_file); goto out; }if (ret & (BDRV_BLOCK_DATA |BDRV_BLOCK_ZERO )) { ret |=BDRV_BLOCK_ALLOCATED ; - }else { + }else if (want_zero) {if (bdrv_unallocated_blocks_are_zero (bs)) { ret |=BDRV_BLOCK_ZERO ; }else if (bs->backing) {BlockDriverState *bs2 = bs->backing->bs;int64_t nb_sectors2 =bdrv_nb_sectors (bs2); +if (nb_sectors2 >=0 && sector_num >= nb_sectors2) { ret |=BDRV_BLOCK_ZERO ; } } } -if (local_file && local_file != bs && +if (want_zero && local_file && local_file != bs && (ret &BDRV_BLOCK_DATA ) && !(ret &BDRV_BLOCK_ZERO ) && (ret &BDRV_BLOCK_OFFSET_VALID )) {int file_pnum; - ret2 =bdrv_co_get_block_status (local_file, ret >>BDRV_SECTOR_BITS , + ret2 =bdrv_co_get_block_status (local_file, want_zero, + ret >>BDRV_SECTOR_BITS , *pnum, &file_pnum,NULL );if (ret2 >=0 ) {/* Ignore errors. This is just providing extra information, it
@@ -1918,6 +1928,7 @@ early_out: static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs, BlockDriverState *base, + bool want_zero, int64_t sector_num, int nb_sectors, int *pnum,
@@ -1929,7 +1940,8 @@ static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,assert (bs != base);for (p = bs; p != base; p =backing_bs (p)) { - ret =bdrv_co_get_block_status (p, sector_num, nb_sectors, pnum, file); + ret =bdrv_co_get_block_status (p, want_zero, sector_num, nb_sectors, + pnum, file);if (ret <0 ) {break ; }
@@ -1959,6 +1971,7 @@ static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)BdrvCoGetBlockStatusData *data = opaque; data->ret = bdrv_co_get_block_status_above(data->bs, data->base, + data->want_zero, data->sector_num, data->nb_sectors, data->pnum,
@@ -1971,20 +1984,22 @@ static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)* * See bdrv_co_get_block_status_above() for details. */ -int64_t bdrv_get_block_status_above (BlockDriverState *bs, -BlockDriverState *base, -int64_t sector_num, -int nb_sectors,int *pnum, -BlockDriverState **file) +static int64_t bdrv_common_block_status_above (BlockDriverState *bs, +BlockDriverState *base, +bool want_zero, +int64_t sector_num, +int nb_sectors,int *pnum, +BlockDriverState **file) {Coroutine *co;BdrvCoGetBlockStatusData data = { .bs = bs, .base = base, - .file = file, + .want_zero = want_zero, .sector_num = sector_num, .nb_sectors = nb_sectors, .pnum = pnum, + .file = file, .done = false, };
@@ -2000,6 +2015,16 @@ int64_t bdrv_get_block_status_above(BlockDriverState *bs,return data.ret; } +int64_t bdrv_get_block_status_above (BlockDriverState *bs, +BlockDriverState *base, +int64_t sector_num, +int nb_sectors,int *pnum, +BlockDriverState **file) +{ +return bdrv_common_block_status_above (bs, base, true, sector_num, + nb_sectors, pnum, file); +} + int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
@@ -2012,15 +2037,15 @@ int64_t bdrv_get_block_status(BlockDriverState *bs, int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset,int64_t bytes,int64_t *pnum) { -int64_t sector_num = offset >>BDRV_SECTOR_BITS ; -int nb_sectors = bytes >>BDRV_SECTOR_BITS ;int64_t ret;int psectors;assert (QEMU_IS_ALIGNED (offset,BDRV_SECTOR_SIZE ));assert (QEMU_IS_ALIGNED (bytes,BDRV_SECTOR_SIZE ) && bytes <INT_MAX ); - ret =bdrv_get_block_status (bs, sector_num, nb_sectors, &psectors, -NULL ); + ret =bdrv_common_block_status_above (bs,backing_bs (bs), false, + offset >>BDRV_SECTOR_BITS , + bytes >>BDRV_SECTOR_BITS , &psectors, +NULL );if (ret <0 ) {return ret; }
Signature · 1 line
2.13.6
PATCH v6 3/24 block: Make bdrv_round_to_clusters() signature more useful
In reply to Eric BlakeIn the process of converting sector-based interfaces to bytes, I'm finding it easier to represent a byte count as a 64-bit integer at the block layer (even if we are internally capped by SIZE_MAX or even INT_MAX for individual transactions, it's still nicer to not have to worry about truncation/overflow issues on as many variables). Update the signature of bdrv_round_to_clusters() to uniformly use int64_t, matching the signature already chosen for bdrv_is_allocated and the fact that off_t is also a signed type, then adjust clients according to the required fallout (even where the result could now exceed 32 bits, no client is directly assigning the result into a 32-bit value without breaking things into a loop first).
---
v5: depends on copy-on-read fixes [John], fix incorrect trace, update commit message to document rounding considerations, drop R-b
v4: only context changes
v3: no change
v2: fix commit message [John], rebase to earlier changes, including mirror_clip_bytes() signature update
--- include/block/block.h | 4 ++-- block/io.c | 6 +++--- block/mirror.c | 7 +++---- block/trace-events | 2 +- 4 files changed, 9 insertions(+), 10 deletions(-)
include/block/block.h +2 -2
@@ -474,9 +474,9 @@ int bdrv_get_flags(BlockDriverState *bs);int bdrv_get_info (BlockDriverState *bs,BlockDriverInfo *bdi);ImageInfoSpecific *bdrv_get_specific_info (BlockDriverState *bs);void bdrv_round_to_clusters (BlockDriverState *bs, -int64_t offset,unsigned int bytes, +int64_t offset,int64_t bytes,int64_t *cluster_offset, -unsigned int *cluster_bytes); +int64_t *cluster_bytes);const char *bdrv_get_encrypted_filename (BlockDriverState *bs); void bdrv_get_backing_filename(BlockDriverState *bs,
block/io.c +3 -3
@@ -449,9 +449,9 @@ static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)* Round a region to cluster boundaries */ void bdrv_round_to_clusters (BlockDriverState *bs, -int64_t offset,unsigned int bytes, +int64_t offset,int64_t bytes,int64_t *cluster_offset, -unsigned int *cluster_bytes) +int64_t *cluster_bytes) {BlockDriverInfo bdi;
@@ -949,7 +949,7 @@ static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child,struct iovec iov;QEMUIOVector local_qiov;int64_t cluster_offset; -unsigned int cluster_bytes; +int64_t cluster_bytes;size_t skip_bytes;int ret; int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer,
block/mirror.c +3 -4
@@ -190,10 +190,9 @@ static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset,bool need_cow;int ret =0 ;int64_t align_offset = *offset; -unsigned int align_bytes = *bytes; +int64_t align_bytes = *bytes;int max_bytes = s->granularity * s->max_iov; -assert (*bytes <INT_MAX ); need_cow = !test_bit (*offset / s->granularity, s->cow_bitmap); need_cow |= !test_bit ((*offset + *bytes -1 ) / s->granularity, s->cow_bitmap);
@@ -388,7 +387,7 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)while (nb_chunks >0 && offset < s->bdev_length) {int64_t ret;int io_sectors; -unsigned int io_bytes; +int64_t io_bytes;int64_t io_bytes_acct;enum MirrorMethod {MIRROR_METHOD_COPY ,
@@ -413,7 +412,7 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) io_bytes = s->granularity; }else if (ret >=0 && !(ret &BDRV_BLOCK_DATA )) {int64_t target_offset; -unsigned int target_bytes; +int64_t target_bytes;bdrv_round_to_clusters (blk_bs (s->target), offset, io_bytes, &target_offset, &target_bytes); if (target_offset == offset &&
block/trace-events +1 -1
@@ -12,7 +12,7 @@ blk_co_pwritev(void *blk, void *bs, int64_t offset, unsigned int bytes, int flag bdrv_co_preadv(void *bs, int64_t offset, int64_t nbytes, unsigned int flags) "bs %p offset %"PRId64" nbytes %"PRId64" flags 0x%x" bdrv_co_pwritev(void *bs, int64_t offset, int64_t nbytes, unsigned int flags) "bs %p offset %"PRId64" nbytes %"PRId64" flags 0x%x" bdrv_co_pwrite_zeroes(void *bs, int64_t offset, int count, int flags) "bs %p offset %"PRId64" count %d flags 0x%x" -bdrv_co_do_copy_on_readv(void *bs, int64_t offset, unsigned int bytes, int64_t cluster_offset, unsigned int cluster_bytes) "bs %p offset %"PRId64" bytes %u cluster_offset %"PRId64" cluster_bytes %u" +bdrv_co_do_copy_on_readv(void *bs, int64_t offset, unsigned int bytes, int64_t cluster_offset, int64_t cluster_bytes) "bs %p offset %"PRId64" bytes %u cluster_offset %"PRId64" cluster_bytes %"PRId64 # block/stream.c stream_one_iteration(void *s, int64_t offset, uint64_t bytes, int is_allocated) "s %p offset %" PRId64 " bytes %" PRIu64 " is_allocated %d"
Signature · 1 line
2.13.6
PATCH v6 5/24 block: Switch bdrv_make_zero() to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Change the internal loop iteration of zeroing a device to track by bytes instead of sectors (although we are still guaranteed that we iterate by steps that are sector-aligned).
---
v3-v5: no change
v2: rebase to earlier changes
--- block/io.c | 32 ++++++++++++++++---------------- 1 file changed, 16 insertions(+), 16 deletions(-)
block/io.c +16 -16
@@ -696,38 +696,38 @@ int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,*/ int bdrv_make_zero (BdrvChild *child,BdrvRequestFlags flags) { -int64_t target_sectors, ret, nb_sectors, sector_num =0 ; +int64_t target_size, ret, bytes, offset =0 ;BlockDriverState *bs = child->bs; -int n; +int n;/* sectors */ - target_sectors =bdrv_nb_sectors (bs); -if (target_sectors <0 ) { -return target_sectors; + target_size =bdrv_getlength (bs); +if (target_size <0 ) { +return target_size; }for (;;) { - nb_sectors =MIN (target_sectors - sector_num,BDRV_REQUEST_MAX_SECTORS ); -if (nb_sectors <=0 ) { + bytes =MIN (target_size - offset,BDRV_REQUEST_MAX_BYTES ); +if (bytes <=0 ) {return 0 ; } - ret =bdrv_get_block_status (bs, sector_num, nb_sectors, &n,NULL ); + ret =bdrv_get_block_status (bs, offset >>BDRV_SECTOR_BITS , + bytes >>BDRV_SECTOR_BITS , &n,NULL );if (ret <0 ) { -error_report ("error getting block status at sector %" PRId64": %s" , - sector_num,strerror (-ret)); +error_report ("error getting block status at offset %" PRId64": %s" , + offset,strerror (-ret));return ret; }if (ret &BDRV_BLOCK_ZERO ) { - sector_num += n; + offset += n *BDRV_SECTOR_BITS ;continue ; } - ret =bdrv_pwrite_zeroes (child, sector_num <<BDRV_SECTOR_BITS , - n <<BDRV_SECTOR_BITS , flags); + ret =bdrv_pwrite_zeroes (child, offset, n *BDRV_SECTOR_SIZE , flags);if (ret <0 ) { -error_report ("error writing zeroes at sector %" PRId64": %s" , - sector_num,strerror (-ret)); +error_report ("error writing zeroes at offset %" PRId64": %s" , + offset,strerror (-ret));return ret; } - sector_num += n; + offset += n *BDRV_SECTOR_SIZE ; } }
Signature · 1 line
2.13.6
PATCH v6 4/24 qcow2: Switch is_zero_sectors() to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Convert another internal function (no semantic change), and rename it to is_zero() in the process.
---
v6: tweak comments to add TODO for later in series [Kevin], minor enough to keep R-b
v3-v5: no change
v2: rename function, rebase to upstream changes
--- block/qcow2.c | 33 +++++++++++++++++++-------------- 1 file changed, 19 insertions(+), 14 deletions(-)
block/qcow2.c +19 -14
@@ -2999,21 +2999,29 @@ finish: } -static bool is_zero_sectors (BlockDriverState *bs,int64_t start, -uint32_t count) +static bool is_zero (BlockDriverState *bs,int64_t offset,int64_t bytes) {int nr;int64_t res; +int64_t start; -if (start + count > bs->total_sectors) { - count = bs->total_sectors - start; +/* TODO: Widening to sector boundaries should only be needed as +* long as we can't query finer granularity. */ + start =QEMU_ALIGN_DOWN (offset,BDRV_SECTOR_SIZE ); + bytes =QEMU_ALIGN_UP (offset + bytes,BDRV_SECTOR_SIZE ) - start; + +/* Clamp to image length, before checking status of underlying sectors */ +if (start + bytes > bs->total_sectors *BDRV_SECTOR_SIZE ) { + bytes = bs->total_sectors *BDRV_SECTOR_SIZE - start; } -if (!count) { +if (!bytes) {return true; } - res =bdrv_get_block_status_above (bs,NULL , start, count, &nr,NULL ); -return res >=0 && (res &BDRV_BLOCK_ZERO ) && nr == count; + res =bdrv_get_block_status_above (bs,NULL , start >>BDRV_SECTOR_BITS , + bytes >>BDRV_SECTOR_BITS , &nr,NULL ); +return res >=0 && (res &BDRV_BLOCK_ZERO ) && + nr *BDRV_SECTOR_SIZE == bytes; } static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
@@ -3031,24 +3039,21 @@ static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, }if (head || tail) { -int64_t cl_start = (offset - head) >>BDRV_SECTOR_BITS ;uint64_t off;unsigned int nr;assert (head + bytes <= s->cluster_size);/* check whether remainder of cluster already reads as zero */ -if (!(is_zero_sectors (bs, cl_start, -DIV_ROUND_UP (head,BDRV_SECTOR_SIZE )) && -is_zero_sectors (bs, (offset + bytes) >>BDRV_SECTOR_BITS , -DIV_ROUND_UP (-tail & (s->cluster_size -1 ), -BDRV_SECTOR_SIZE )))) { +if (!(is_zero (bs, offset - head, head) && +is_zero (bs, offset + bytes, + tail ? s->cluster_size - tail :0 ))) {return -ENOTSUP ; }qemu_co_mutex_lock (&s->lock);/* We can have new write after previous check */ - offset = cl_start <<BDRV_SECTOR_BITS ; + offset =QEMU_ALIGN_DOWN (offset, s->cluster_size); bytes = s->cluster_size; nr = s->cluster_size; ret =qcow2_get_cluster_offset (bs, offset, &nr, &off);
Signature · 1 line
2.13.6
PATCH v6 6/24 qemu-img: Switch get_block_status() to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Continue by converting an internal function (no semantic change), and simplifying its caller accordingly.
---
v2-v5: no change
--- qemu-img.c | 24 +++++++++++------------- 1 file changed, 11 insertions(+), 13 deletions(-)
qemu-img.c +11 -13
@@ -2671,14 +2671,16 @@ static void dump_map_entry(OutputFormat output_format, MapEntry *e, } } -static int get_block_status (BlockDriverState *bs,int64_t sector_num, -int nb_sectors,MapEntry *e) +static int get_block_status (BlockDriverState *bs,int64_t offset, +int64_t bytes,MapEntry *e) {int64_t ret;int depth;BlockDriverState *file;bool has_offset; +int nb_sectors = bytes >>BDRV_SECTOR_BITS ; +assert (bytes <INT_MAX );/* As an optimization, we could cache the current range of unallocated * clusters in each file of the chain, and avoid querying the same * range repeatedly.
@@ -2686,8 +2688,8 @@ static int get_block_status(BlockDriverState *bs, int64_t sector_num, depth =0 ;for (;;) { - ret =bdrv_get_block_status (bs, sector_num, nb_sectors, &nb_sectors, - &file); + ret =bdrv_get_block_status (bs, offset >>BDRV_SECTOR_BITS , nb_sectors, + &nb_sectors, &file);if (ret <0 ) {return ret; }
@@ -2707,7 +2709,7 @@ static int get_block_status(BlockDriverState *bs, int64_t sector_num, has_offset = !!(ret &BDRV_BLOCK_OFFSET_VALID ); *e = (MapEntry ) { - .start = sector_num *BDRV_SECTOR_SIZE , + .start = offset, .length = nb_sectors *BDRV_SECTOR_SIZE , .data = !!(ret &BDRV_BLOCK_DATA ), .zero = !!(ret &BDRV_BLOCK_ZERO ),
@@ -2837,16 +2839,12 @@ static int img_map(int argc, char **argv) length =blk_getlength (blk);while (curr.start + curr.length < length) { -int64_t nsectors_left; -int64_t sector_num; -int n; - - sector_num = (curr.start + curr.length) >>BDRV_SECTOR_BITS ; +int64_t offset = curr.start + curr.length; +int64_t n;/* Probe up to 1 GiB at a time. */ - nsectors_left =DIV_ROUND_UP (length,BDRV_SECTOR_SIZE ) - sector_num; - n =MIN (1 << (30 -BDRV_SECTOR_BITS ), nsectors_left); - ret =get_block_status (bs, sector_num, n, &next); + n =QEMU_ALIGN_DOWN (MIN (1 <<30 , length - offset),BDRV_SECTOR_SIZE ); + ret =get_block_status (bs, offset, n, &next);if (ret <0 ) {error_report ("Could not read file metadata: %s" ,strerror (-ret));
Signature · 1 line
2.13.6
PATCH v6 7/24 block: Convert bdrv_get_block_status() to bytes
In reply to Eric BlakeWe are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers.
There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses.
For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors.
For ease of review, bdrv_get_block_status_above() will be tackled separately.
---
v6: redo interface to return byte-based offset by reference [Kevin]
v5: drop pointless 'if (pnum)' [John], add comment
v4: no change
v3: clamp bytes to 32-bits, rather than asserting
v2: rebase to earlier changes
--- include/block/block.h | 17 +++++++++-------- block/io.c | 47 ++++++++++++++++++++++++++++++++++------------- block/qcow2-cluster.c | 2 +- qemu-img.c | 25 ++++++++++++++----------- 4 files changed, 58 insertions(+), 33 deletions(-)
include/block/block.h +9 -8
@@ -121,7 +121,7 @@ typedef struct HDGeometry {#define BDRV_REQUEST_MAX_BYTES (BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS)/* -* Allocation status flags for bdrv_get_block_status() and friends. +* Allocation status flags for bdrv_block_status() and friends. * * Public flags: * BDRV_BLOCK_DATA: allocation for data at offset is tied to this layer
@@ -136,10 +136,11 @@ typedef struct HDGeometry {* that the block layer recompute the answer from the returned * BDS; must be accompanied by just BDRV_BLOCK_OFFSET_VALID. * -* If BDRV_BLOCK_OFFSET_VALID is set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK) -* represent the offset in the returned BDS that is allocated for the -* corresponding raw data; however, whether that offset actually contains -* data also depends on BDRV_BLOCK_DATA and BDRV_BLOCK_ZERO, as follows: +* If BDRV_BLOCK_OFFSET_VALID is set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK) of +* the return value (old interface) or the entire map parameter (new +* interface) represent the offset in the returned BDS that is allocated for +* the corresponding raw data. However, whether that offset actually +* contains data also depends on BDRV_BLOCK_DATA, as follows: * * DATA ZERO OFFSET_VALID * t t t sectors read as zero, returned file is zero at offset
@@ -421,9 +422,9 @@ int bdrv_has_zero_init_1(BlockDriverState *bs);int bdrv_has_zero_init (BlockDriverState *bs);bool bdrv_unallocated_blocks_are_zero (BlockDriverState *bs);bool bdrv_can_write_zeroes_with_unmap (BlockDriverState *bs); -int64_t bdrv_get_block_status (BlockDriverState *bs,int64_t sector_num, -int nb_sectors,int *pnum, -BlockDriverState **file); +int bdrv_block_status (BlockDriverState *bs,int64_t offset, +int64_t bytes,int64_t *pnum,int64_t *map, +BlockDriverState **file); int64_t bdrv_get_block_status_above(BlockDriverState *bs, BlockDriverState *base, int64_t sector_num,
block/io.c +34 -13
80 lines not on this page · raw message
block/qcow2-cluster.c +1 -1
@@ -1632,7 +1632,7 @@ static int discard_single_l2(BlockDriverState *bs, uint64_t offset,* cluster is already marked as zero, or if it's unallocated and we * don't have a backing file. * -* TODO We might want to use bdrv_get_block_status(bs) here, but we're +* TODO We might want to use bdrv_block_status(bs) here, but we're * holding s->lock, so that doesn't work today. * * If full_discard is true, the sector should not read back as zeroes,
qemu-img.c +14 -11
@@ -1599,9 +1599,14 @@ static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)if (s->sector_next_status <= sector_num) {if (s->target_has_backing) { - ret =bdrv_get_block_status (blk_bs (s->src[src_cur]), - sector_num - src_cur_offset, - n, &n,NULL ); +int64_t count = n *BDRV_SECTOR_SIZE ; + + ret =bdrv_block_status (blk_bs (s->src[src_cur]), + (sector_num - src_cur_offset) * +BDRV_SECTOR_SIZE , + count, &count,NULL ,NULL ); +assert (ret <0 ||QEMU_IS_ALIGNED (count,BDRV_SECTOR_SIZE )); + n = count >>BDRV_SECTOR_BITS ; }else { ret = bdrv_get_block_status_above(blk_bs(s->src[src_cur]), NULL, sector_num - src_cur_offset,
@@ -2674,13 +2679,12 @@ static void dump_map_entry(OutputFormat output_format, MapEntry *e,static int get_block_status (BlockDriverState *bs,int64_t offset,int64_t bytes,MapEntry *e) { -int64_t ret; +int ret;int depth;BlockDriverState *file;bool has_offset; -int nb_sectors = bytes >>BDRV_SECTOR_BITS ; +int64_t map; -assert (bytes <INT_MAX );/* As an optimization, we could cache the current range of unallocated * clusters in each file of the chain, and avoid querying the same * range repeatedly.
@@ -2688,12 +2692,11 @@ static int get_block_status(BlockDriverState *bs, int64_t offset, depth =0 ;for (;;) { - ret =bdrv_get_block_status (bs, offset >>BDRV_SECTOR_BITS , nb_sectors, - &nb_sectors, &file); + ret =bdrv_block_status (bs, offset, bytes, &bytes, &map, &file);if (ret <0 ) {return ret; } -assert (nb_sectors); +assert (bytes);if (ret & (BDRV_BLOCK_ZERO |BDRV_BLOCK_DATA )) {break ; }
@@ -2710,10 +2713,10 @@ static int get_block_status(BlockDriverState *bs, int64_t offset, *e = (MapEntry ) { .start = offset, - .length = nb_sectors *BDRV_SECTOR_SIZE , + .length = bytes, .data = !!(ret &BDRV_BLOCK_DATA ), .zero = !!(ret &BDRV_BLOCK_ZERO ), - .offset = ret &BDRV_BLOCK_OFFSET_MASK , + .offset = map, .has_offset = has_offset, .depth = depth, .has_filename = file && has_offset,
Signature · 1 line
2.13.6
PATCH v6 8/24 block: Switch bdrv_co_get_block_status() to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Convert another internal function (no semantic change); and as with its public counterpart, rename to bdrv_co_block_status() and split the offset return, to make the compiler enforce that we catch all uses. For now, we assert that callers and the return value still use aligned data, but ultimately, this will be the function where we hand off to a byte-based driver callback, and will eventually need to add logic to ensure we round calls according to the driver's request_alignment then touch up the result handed back to the caller, to start permitting a caller to pass unaligned offsets.
Note that we are now prepared to accepts 'bytes' larger than INT_MAX; this is okay as long as we clamp things internally before violating any 32-bit limits, and makes no difference to how a client will use the information (clients looping over the entire file must already be prepared for consecutive calls to return the same status, as drivers are already free to return shorter-than-maximal status due to any other convenient split points, such as when the L2 table crosses cluster boundaries in qcow2).
---
v6: rebase to return value interface change [Kevin]
v5: rebase to earlier changes in 1/23, add comment
v4: no change
v3: rebase to allocation/mapping sense change, clamp bytes to 32-bits when needed, drop R-b
v2: rebase to earlier changes
--- block/io.c | 122 ++++++++++++++++++++++++++++++++++++++++--------------------- 1 file changed, 80 insertions(+), 42 deletions(-)
block/io.c +80 -42
@@ -1793,80 +1793,102 @@ int64_t coroutine_fn bdrv_co_get_block_status_from_backing(BlockDriverState *bs,* BDRV_BLOCK_ZERO where possible; otherwise, the result may omit those * bits particularly if it allows for a larger value in 'pnum'. * -* If 'sector_num' is beyond the end of the disk image the return value is +* If 'offset' is beyond the end of the disk image the return value is * BDRV_BLOCK_EOF and 'pnum' is set to 0. * -* 'pnum' is set to the number of sectors (including and immediately following -* the specified sector) that are known to be in the same -* allocated/unallocated state. -* -* 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes +* 'bytes' is the max value 'pnum' should be set to. If bytes goes * beyond the end of the disk image it will be clamped; if 'pnum' is set to * the end of the image, then the returned value will include BDRV_BLOCK_EOF. * -* If returned value is positive, BDRV_BLOCK_OFFSET_VALID bit is set, and -* 'file' is non-NULL, then '*file' points to the BDS which the sector range -* is allocated in. +* 'pnum' is set to the number of bytes (including and immediately +* following the specified offset) that are easily known to be in the +* same allocated/unallocated state. Note that a second call starting +* at the original offset plus returned pnum may have the same status. +* The returned value is non-zero on success except at end-of-file. +* +* Returns negative errno on failure. Otherwise, if the +* BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are +* set to the host mapping and BDS corresponding to the guest offset. */ -static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, -bool want_zero, -int64_t sector_num, -int nb_sectors,int *pnum, -BlockDriverState **file) +static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs, +bool want_zero, +int64_t offset,int64_t bytes, +int64_t *pnum,int64_t *map, +BlockDriverState **file) { -int64_t total_sectors; -int64_t n; -int64_t ret, ret2; +int64_t total_size; +int64_t n;/* bytes */ +int64_t ret; +int64_t local_map =0 ;BlockDriverState *local_file =NULL ; +int count;/* sectors */ assert (pnum); *pnum =0 ; - total_sectors =bdrv_nb_sectors (bs); -if (total_sectors <0 ) { - ret = total_sectors; + total_size =bdrv_getlength (bs); +if (total_size <0 ) { + ret = total_size; goto early_out; } -if (sector_num >= total_sectors) { +if (offset >= total_size) { ret =BDRV_BLOCK_EOF ; goto early_out; } -if (!nb_sectors) { +if (!bytes) { ret =0 ; goto early_out; } - n = total_sectors - sector_num; -if (n < nb_sectors) { - nb_sectors = n; + n = total_size - offset; +if (n < bytes) { + bytes = n; }if (!bs->drv->bdrv_co_get_block_status) { - *pnum = nb_sectors; + *pnum = bytes; ret =BDRV_BLOCK_DATA |BDRV_BLOCK_ALLOCATED ; -if (sector_num + nb_sectors == total_sectors) { +if (offset + bytes == total_size) { ret |=BDRV_BLOCK_EOF ; }if (bs->drv->protocol_name) { - ret |=BDRV_BLOCK_OFFSET_VALID | (sector_num *BDRV_SECTOR_SIZE ); + ret |=BDRV_BLOCK_OFFSET_VALID ; + local_map = offset; local_file = bs; } goto early_out; }bdrv_inc_in_flight (bs); - ret = bs->drv->bdrv_co_get_block_status (bs, sector_num, nb_sectors, pnum, +/* +* TODO: Rather than require aligned offsets, we could instead +* round to the driver's request_alignment here, then touch up +* count afterwards back to the caller's expectations. +*/ +assert (QEMU_IS_ALIGNED (offset | bytes,BDRV_SECTOR_SIZE )); +/* +* The contract allows us to return pnum smaller than bytes, even +* if the next query would see the same status; we truncate the +* request to avoid overflowing the driver's 32-bit interface. +*/ + bytes =MIN (bytes,BDRV_REQUEST_MAX_BYTES ); + ret = bs->drv->bdrv_co_get_block_status (bs, offset >>BDRV_SECTOR_BITS , + bytes >>BDRV_SECTOR_BITS , &count, &local_file);if (ret <0 ) { - *pnum =0 ; goto out; } +if (ret &BDRV_BLOCK_OFFSET_VALID ) { + local_map = ret &BDRV_BLOCK_OFFSET_MASK ; + } + *pnum = count *BDRV_SECTOR_SIZE ;if (ret &BDRV_BLOCK_RAW ) {assert (ret &BDRV_BLOCK_OFFSET_VALID && local_file); - ret =bdrv_co_get_block_status (local_file, want_zero, - ret >>BDRV_SECTOR_BITS , - *pnum, pnum, &local_file); + ret =bdrv_co_block_status (local_file, want_zero, local_map, + *pnum, pnum, &local_map, &local_file); +assert (ret <0 || +QEMU_IS_ALIGNED (*pnum | local_map,BDRV_SECTOR_SIZE )); goto out; }
@@ -1877,9 +1899,9 @@ static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, ret |=BDRV_BLOCK_ZERO ; }else if (bs->backing) {BlockDriverState *bs2 = bs->backing->bs; -int64_t nb_sectors2 =bdrv_nb_sectors (bs2); +int64_t size2 =bdrv_getlength (bs2); -if (nb_sectors2 >=0 && sector_num >= nb_sectors2) { +if (size2 >=0 && offset >= size2) { ret |=BDRV_BLOCK_ZERO ; } }
@@ -1888,11 +1910,11 @@ static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,if (want_zero && local_file && local_file != bs && (ret &BDRV_BLOCK_DATA ) && !(ret &BDRV_BLOCK_ZERO ) && (ret &BDRV_BLOCK_OFFSET_VALID )) { -int file_pnum; +int64_t file_pnum; +int ret2; - ret2 =bdrv_co_get_block_status (local_file, want_zero, - ret >>BDRV_SECTOR_BITS , - *pnum, &file_pnum,NULL ); + ret2 =bdrv_co_block_status (local_file, want_zero, local_map, + *pnum, &file_pnum,NULL ,NULL );if (ret2 >=0 ) {/* Ignore errors. This is just providing extra information, it * is useful but not necessary.
@@ -1915,13 +1937,21 @@ static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, out:bdrv_dec_in_flight (bs); -if (ret >=0 && sector_num + *pnum == total_sectors) { +if (ret >=0 && offset + *pnum == total_size) { ret |=BDRV_BLOCK_EOF ; } early_out:if (file) { *file = local_file; } +if (map) { + *map = local_map; + } +if (ret >=0 ) { + ret &= ~BDRV_BLOCK_OFFSET_MASK ; + }else { +assert (INT_MIN <= ret); + }return ret; }
@@ -1936,14 +1966,22 @@ static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,BlockDriverState *p;int64_t ret =0 ;bool first = true; +int64_t map =0 ;assert (bs != base);for (p = bs; p != base; p =backing_bs (p)) { - ret =bdrv_co_get_block_status (p, want_zero, sector_num, nb_sectors, - pnum, file); +int64_t count; + + ret =bdrv_co_block_status (p, want_zero, + sector_num *BDRV_SECTOR_SIZE , + nb_sectors *BDRV_SECTOR_SIZE , &count, + &map, file);if (ret <0 ) {break ; } +assert (QEMU_IS_ALIGNED (count | map,BDRV_SECTOR_SIZE )); + ret |= map; + *pnum = count >>BDRV_SECTOR_BITS ;if (ret &BDRV_BLOCK_ZERO && ret &BDRV_BLOCK_EOF && !first) {/* * Reading beyond the end of the file continues to read
Signature · 1 line
2.13.6
PATCH v6 9/24 block: Switch BdrvCoGetBlockStatusData to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Convert another internal type (no semantic change), and rename it to match the corresponding public function rename.
---
v6: fix -O2 compilation [Jeff], rebase to split return interface change [Kevin], drop R-b
v4-v5: no change
v3: rebase to context conflicts, simple enough to keep R-b
v2: rebase to earlier changes
--- block/io.c | 56 ++++++++++++++++++++++++++++++++++++++------------------ 1 file changed, 38 insertions(+), 18 deletions(-)
block/io.c +38 -18
@@ -1745,17 +1745,18 @@ int bdrv_flush_all(void) } -typedef struct BdrvCoGetBlockStatusData { +typedef struct BdrvCoBlockStatusData {BlockDriverState *bs;BlockDriverState *base;bool want_zero; -int64_t sector_num; -int nb_sectors; -int *pnum; +int64_t offset; +int64_t bytes; +int64_t *pnum; +int64_t *map;BlockDriverState **file; -int64_t ret; +int ret;bool done; -}BdrvCoGetBlockStatusData ; +}BdrvCoBlockStatusData ; int64_t coroutine_fn bdrv_co_get_block_status_from_file(BlockDriverState *bs, int64_t sector_num,
@@ -2005,14 +2006,24 @@ static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,/* Coroutine wrapper for bdrv_get_block_status_above() */ static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque) { -BdrvCoGetBlockStatusData *data = opaque; +BdrvCoBlockStatusData *data = opaque; +int n =0 ; +int64_t ret; - data->ret =bdrv_co_get_block_status_above (data->bs, data->base, - data->want_zero, - data->sector_num, - data->nb_sectors, - data->pnum, - data->file); + ret =bdrv_co_get_block_status_above (data->bs, data->base, + data->want_zero, + data->offset >>BDRV_SECTOR_BITS , + data->bytes >>BDRV_SECTOR_BITS , + &n, + data->file); +if (ret <0 ) { +assert (INT_MIN <= ret); + data->ret = ret; + }else { + *data->pnum = n *BDRV_SECTOR_SIZE ; + *data->map = ret &BDRV_BLOCK_OFFSET_MASK ; + data->ret = ret & ~BDRV_BLOCK_OFFSET_MASK ; + } data->done = true; }
@@ -2029,13 +2040,16 @@ static int64_t bdrv_common_block_status_above(BlockDriverState *bs, BlockDriverState **file) {Coroutine *co; -BdrvCoGetBlockStatusData data = { +int64_t n; +int64_t map; +BdrvCoBlockStatusData data = { .bs = bs, .base = base, .want_zero = want_zero, - .sector_num = sector_num, - .nb_sectors = nb_sectors, - .pnum = pnum, + .offset = sector_num *BDRV_SECTOR_SIZE , + .bytes = nb_sectors *BDRV_SECTOR_SIZE , + .pnum = &n, + .map = &map, .file = file, .done = false, };
@@ -2049,7 +2063,13 @@ static int64_t bdrv_common_block_status_above(BlockDriverState *bs,bdrv_coroutine_enter (bs, co);BDRV_POLL_WHILE (bs, !data.done); } -return data.ret; +if (data.ret <0 ) { + *pnum =0 ; +return data.ret; + } +assert (QEMU_IS_ALIGNED (n | map,BDRV_SECTOR_SIZE )); + *pnum = n >>BDRV_SECTOR_BITS ; +return data.ret | map; } int64_t bdrv_get_block_status_above(BlockDriverState *bs,
Signature · 1 line
2.13.6
PATCH v6 10/24 block: Switch bdrv_common_block_status_above() to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Convert another internal function (no semantic change).
---
v6: rebase to split return interface change [Kevin], R-b dropped
v4-v5: no change
v3: rebase to allocation/mapping sense change, simple enough to keep R-b
v2: new patch
--- block/io.c | 61 ++++++++++++++++++++++++++++++------------------------------- 1 file changed, 30 insertions(+), 31 deletions(-)
block/io.c +30 -31
@@ -2032,24 +2032,22 @@ static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)* * See bdrv_co_get_block_status_above() for details. */ -static int64_t bdrv_common_block_status_above (BlockDriverState *bs, -BlockDriverState *base, -bool want_zero, -int64_t sector_num, -int nb_sectors,int *pnum, -BlockDriverState **file) +static int bdrv_common_block_status_above (BlockDriverState *bs, +BlockDriverState *base, +bool want_zero,int64_t offset, +int64_t bytes,int64_t *pnum, +int64_t *map, +BlockDriverState **file) {Coroutine *co; -int64_t n; -int64_t map;BdrvCoBlockStatusData data = { .bs = bs, .base = base, .want_zero = want_zero, - .offset = sector_num *BDRV_SECTOR_SIZE , - .bytes = nb_sectors *BDRV_SECTOR_SIZE , - .pnum = &n, - .map = &map, + .offset = offset, + .bytes = bytes, + .pnum = pnum, + .map = map, .file = file, .done = false, };
@@ -2063,13 +2061,7 @@ static int64_t bdrv_common_block_status_above(BlockDriverState *bs,bdrv_coroutine_enter (bs, co);BDRV_POLL_WHILE (bs, !data.done); } -if (data.ret <0 ) { - *pnum =0 ; -return data.ret; - } -assert (QEMU_IS_ALIGNED (n | map,BDRV_SECTOR_SIZE )); - *pnum = n >>BDRV_SECTOR_BITS ; -return data.ret | map; +return data.ret; } int64_t bdrv_get_block_status_above(BlockDriverState *bs,
@@ -2078,8 +2070,21 @@ int64_t bdrv_get_block_status_above(BlockDriverState *bs, int nb_sectors, int *pnum, BlockDriverState **file) { - return bdrv_common_block_status_above(bs, base, true, sector_num, -nb_sectors ,pnum ,file ); + int64_t ret; +int64_t n; +int64_t map; + + ret =bdrv_common_block_status_above (bs, base, true, + sector_num *BDRV_SECTOR_SIZE , + nb_sectors *BDRV_SECTOR_SIZE , + &n, &map, file); +if (ret <0 ) { + *pnum =0 ; +return ret; + } +assert (QEMU_IS_ALIGNED (n | map,BDRV_SECTOR_SIZE )); + *pnum = n >>BDRV_SECTOR_BITS ; +return ret | map; } int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes,
@@ -2116,21 +2121,15 @@ int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes, int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset,int64_t bytes,int64_t *pnum) { -int64_t ret; -int psectors; +int ret; +int64_t dummy; -assert (QEMU_IS_ALIGNED (offset,BDRV_SECTOR_SIZE )); -assert (QEMU_IS_ALIGNED (bytes,BDRV_SECTOR_SIZE ) && bytes <INT_MAX ); - ret =bdrv_common_block_status_above (bs,backing_bs (bs), false, - offset >>BDRV_SECTOR_BITS , - bytes >>BDRV_SECTOR_BITS , &psectors, + ret =bdrv_common_block_status_above (bs,backing_bs (bs), false, offset, + bytes, pnum ? pnum : &dummy,NULL ,NULL );if (ret <0 ) {return ret; } -if (pnum) { - *pnum = psectors *BDRV_SECTOR_SIZE ; - }return !!(ret &BDRV_BLOCK_ALLOCATED ); }
Signature · 1 line
2.13.6
PATCH v6 11/24 block: Switch bdrv_co_get_block_status_above() to byte-based
In reply to Eric BlakeWe are gradually converting to byte-based interfaces, as they are easier to reason about than sector-based. Convert another internal type (no semantic change), and rename it to match the corresponding public function rename.
---
v6: rebase to split return interface change [Kevin], R-b dropped
v5: no change other than rebase to context
v4: no change
v3: rebase to allocation/mapping sense change, simple enough to keep R-b
v2: rebase to earlier changes
--- block/io.c | 68 ++++++++++++++++++++++---------------------------------------- 1 file changed, 24 insertions(+), 44 deletions(-)
block/io.c +24 -44
@@ -1956,33 +1956,26 @@ early_out:return ret; } -static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs, -BlockDriverState *base, -bool want_zero, -int64_t sector_num, -int nb_sectors, -int *pnum, -BlockDriverState **file) +static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, +BlockDriverState *base, +bool want_zero, +int64_t offset, +int64_t bytes, +int64_t *pnum, +int64_t *map, +BlockDriverState **file) {BlockDriverState *p; -int64_t ret =0 ; +int ret =0 ;bool first = true; -int64_t map =0 ;assert (bs != base);for (p = bs; p != base; p =backing_bs (p)) { -int64_t count; - - ret =bdrv_co_block_status (p, want_zero, - sector_num *BDRV_SECTOR_SIZE , - nb_sectors *BDRV_SECTOR_SIZE , &count, - &map, file); + ret =bdrv_co_block_status (p, want_zero, offset, bytes, pnum, map, + file);if (ret <0 ) {break ; } -assert (QEMU_IS_ALIGNED (count | map,BDRV_SECTOR_SIZE )); - ret |= map; - *pnum = count >>BDRV_SECTOR_BITS ;if (ret &BDRV_BLOCK_ZERO && ret &BDRV_BLOCK_EOF && !first) {/* * Reading beyond the end of the file continues to read
@@ -1990,47 +1983,35 @@ static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,* unallocated length we learned from an earlier * iteration. */ - *pnum = nb_sectors; + *pnum = bytes; }if (ret & (BDRV_BLOCK_ZERO |BDRV_BLOCK_DATA )) {break ; } -/* [sector_num, pnum] unallocated on this layer, which could be only -* the first part of [sector_num, nb_sectors]. */ - nb_sectors =MIN (nb_sectors, *pnum); +/* [offset, pnum] unallocated on this layer, which could be only +* the first part of [offset, bytes]. */ + bytes =MIN (bytes, *pnum); first = false; }return ret; }/* Coroutine wrapper for bdrv_get_block_status_above() */ -static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque) +static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque) {BdrvCoBlockStatusData *data = opaque; -int n =0 ; -int64_t ret; - ret =bdrv_co_get_block_status_above (data->bs, data->base, - data->want_zero, - data->offset >>BDRV_SECTOR_BITS , - data->bytes >>BDRV_SECTOR_BITS , - &n, - data->file); -if (ret <0 ) { -assert (INT_MIN <= ret); - data->ret = ret; - }else { - *data->pnum = n *BDRV_SECTOR_SIZE ; - *data->map = ret &BDRV_BLOCK_OFFSET_MASK ; - data->ret = ret & ~BDRV_BLOCK_OFFSET_MASK ; - } + data->ret =bdrv_co_block_status_above (data->bs, data->base, + data->want_zero, + data->offset, data->bytes, + data->pnum, data->map, data->file); data->done = true; }/* -* Synchronous wrapper around bdrv_co_get_block_status_above(). +* Synchronous wrapper around bdrv_co_block_status_above(). * -* See bdrv_co_get_block_status_above() for details. +* See bdrv_co_block_status_above() for details. */ static int bdrv_common_block_status_above(BlockDriverState *bs, BlockDriverState *base,
@@ -2054,10 +2035,9 @@ static int bdrv_common_block_status_above(BlockDriverState *bs,if (qemu_in_coroutine ()) {/* Fast-path if already in coroutine context */ -bdrv_get_block_status_above_co_entry (&data); +bdrv_block_status_above_co_entry (&data); }else { - co =qemu_coroutine_create (bdrv_get_block_status_above_co_entry, - &data); + co =qemu_coroutine_create (bdrv_block_status_above_co_entry, &data);bdrv_coroutine_enter (bs, co);BDRV_POLL_WHILE (bs, !data.done); }
Signature · 1 line
2.13.6
PATCH v6 12/24 block: Convert bdrv_get_block_status_above() to bytes
In reply to Eric BlakeWe are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers.
For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there.
For ease of review, bdrv_get_block_status() was tackled separately.
---
v6: rebase to split return interface change [Kevin]
v5: assert alignment rather than rounding up in img_compare [John], rebase to earlier changes
v4: rebase to earlier changes
v3: rebase to allocation/mapping sense change and qcow2-measure, tweak mirror assertions, drop R-b
v2: rebase to earlier changes
--- include/block/block.h | 8 +++----- block/io.c | 55 ++++++++------------------------------------------- block/mirror.c | 18 ++++++----------- block/qcow2.c | 30 +++++++++++----------------- qemu-img.c | 49 +++++++++++++++++++++++++-------------------- 5 files changed, 57 insertions(+), 103 deletions(-)
include/block/block.h +3 -5
@@ -425,11 +425,9 @@ bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs);int bdrv_block_status (BlockDriverState *bs,int64_t offset,int64_t bytes,int64_t *pnum,int64_t *map,BlockDriverState **file); -int64_t bdrv_get_block_status_above (BlockDriverState *bs, -BlockDriverState *base, -int64_t sector_num, -int nb_sectors,int *pnum, -BlockDriverState **file); +int bdrv_block_status_above (BlockDriverState *bs,BlockDriverState *base, +int64_t offset,int64_t bytes,int64_t *pnum, +int64_t *map,BlockDriverState **file);int bdrv_is_allocated (BlockDriverState *bs,int64_t offset,int64_t bytes,int64_t *pnum); int bdrv_is_allocated_above(BlockDriverState *top, BlockDriverState *base,
block/io.c +8 -47
73 lines not on this page · raw message
block/mirror.c +6 -12
@@ -328,7 +328,6 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)uint64_t delay_ns =0 ;/* At least the first dirty chunk is mirrored in one iteration. */ int nb_chunks =1 ; -int sectors_per_chunk = s->granularity >>BDRV_SECTOR_BITS ;bool write_zeroes_ok =bdrv_can_write_zeroes_with_unmap (blk_bs (s->target));int max_io_bytes =MAX (s->buf_size /MAX_IN_FLIGHT ,MAX_IO_BYTES );
@@ -376,7 +375,7 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) }/* Clear dirty bits before querying the block status, because -* calling bdrv_get_block_status_above could yield - if some blocks are +* calling bdrv_block_status_above could yield - if some blocks are * marked dirty in this window, we need to know. */ bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset,
@@ -385,8 +384,7 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)bitmap_set (s->in_flight_bitmap, offset / s->granularity, nb_chunks);while (nb_chunks >0 && offset < s->bdev_length) { -int64_t ret; -int io_sectors; +int ret;int64_t io_bytes;int64_t io_bytes_acct; enum MirrorMethod {
@@ -396,11 +394,9 @@ static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) } mirror_method =MIRROR_METHOD_COPY ;assert (!(offset % s->granularity)); - ret =bdrv_get_block_status_above (source,NULL , - offset >>BDRV_SECTOR_BITS , - nb_chunks * sectors_per_chunk, - &io_sectors,NULL ); - io_bytes = io_sectors *BDRV_SECTOR_SIZE ; + ret =bdrv_block_status_above (source,NULL , offset, + nb_chunks * s->granularity, + &io_bytes,NULL ,NULL );if (ret <0 ) { io_bytes =MIN (nb_chunks * s->granularity, max_io_bytes); }else if (ret &BDRV_BLOCK_DATA ) {
@@ -1131,9 +1127,7 @@ static void mirror_start_job(const char *job_id, BlockDriverState *bs, granularity =bdrv_get_default_bitmap_granularity (target); } -assert ((granularity & (granularity -1 )) ==0 ); -/* Granularity must be large enough for sector-based dirty bitmap */ -assert (granularity >=BDRV_SECTOR_SIZE ); +assert (is_power_of_2 (granularity));if (buf_size <0 ) {error_setg (errp,"Invalid parameter 'buf-size'" );
block/qcow2.c +12 -18
@@ -3001,8 +3001,8 @@ finish:static bool is_zero (BlockDriverState *bs,int64_t offset,int64_t bytes) { -int nr; -int64_t res; +int64_t nr; +int res;int64_t start;/* TODO: Widening to sector boundaries should only be needed as
@@ -3018,10 +3018,8 @@ static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)if (!bytes) {return true; } - res =bdrv_get_block_status_above (bs,NULL , start >>BDRV_SECTOR_BITS , - bytes >>BDRV_SECTOR_BITS , &nr,NULL ); -return res >=0 && (res &BDRV_BLOCK_ZERO ) && - nr *BDRV_SECTOR_SIZE == bytes; + res =bdrv_block_status_above (bs,NULL , start, bytes, &nr,NULL ,NULL ); +return res >=0 && (res &BDRV_BLOCK_ZERO ) && nr == bytes; } static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
@@ -3729,17 +3727,14 @@ static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, required = virtual_size; }else {int64_t offset; -int pnum =0 ; +int64_t pnum =0 ; -for (offset =0 ; offset < ssize; - offset += pnum *BDRV_SECTOR_SIZE ) { -int nb_sectors =MIN (ssize - offset, -BDRV_REQUEST_MAX_BYTES ) /BDRV_SECTOR_SIZE ; -int64_t ret; +for (offset =0 ; offset < ssize; offset += pnum) { +int ret; - ret =bdrv_get_block_status_above (in_bs,NULL , - offset >>BDRV_SECTOR_BITS , - nb_sectors, &pnum,NULL ); + ret =bdrv_block_status_above (in_bs,NULL , offset, + ssize - offset, &pnum,NULL , +NULL );if (ret <0 ) {error_setg_errno (&local_err, -ret,"Unable to get block status" );
@@ -3751,11 +3746,10 @@ static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, }else if ((ret & (BDRV_BLOCK_DATA |BDRV_BLOCK_ALLOCATED )) == (BDRV_BLOCK_DATA |BDRV_BLOCK_ALLOCATED )) {/* Extend pnum to end of cluster for next iteration */ - pnum = (ROUND_UP (offset + pnum *BDRV_SECTOR_SIZE , - cluster_size) - offset) >>BDRV_SECTOR_BITS ; + pnum =ROUND_UP (offset + pnum, cluster_size) - offset;/* Count clusters we've seen */ - required += offset % cluster_size + pnum *BDRV_SECTOR_SIZE ; + required += offset % cluster_size + pnum; } } }
qemu-img.c +28 -21
113 lines not on this page · raw message
Signature · 1 line
2.13.6
PATCH v6 13/24 qemu-img: Simplify logic in img_compare()
In reply to Eric BlakeAs long as we are querying the status for a chunk smaller than the known image size, we are guaranteed that a successful return will have set pnum to a non-zero size (pnum is zero only for queries beyond the end of the file). Use that to slightly simplify the calculation of the current chunk size being compared. Likewise, we don't have to shrink the amount of data operated on until we know we have to read the file, and therefore have to fit in the bounds of our buffer. Also, note that 'total_sectors_over' is equivalent to 'progress_base'.
With these changes in place, sectors_to_process() is now dead code, and can be removed.
---
v6: rebase to context
v5: rebase to alignment assertion [John]
v4: no change
v3: new patch
--- qemu-img.c | 38 +++++++++++--------------------------- 1 file changed, 11 insertions(+), 27 deletions(-)
qemu-img.c +11 -27
@@ -1172,11 +1172,6 @@ static int64_t sectors_to_bytes(int64_t sectors)return sectors <<BDRV_SECTOR_BITS ; } -static int64_t sectors_to_process (int64_t total,int64_t from) -{ -return MIN (total - from,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); -} -/* * Check if passed sectors are empty (not allocated or contain only 0 bytes) *
@@ -1373,13 +1368,9 @@ static int img_compare(int argc, char **argv) goto out; } -for (;;) { +while (sector_num < total_sectors) {int status1, status2; - nb_sectors =sectors_to_process (total_sectors, sector_num); -if (nb_sectors <=0 ) { -break ; - } status1 = bdrv_block_status_above(bs1, NULL, sector_num * BDRV_SECTOR_SIZE, (total_sectors1 - sector_num) *
@@ -1406,12 +1397,9 @@ static int img_compare(int argc, char **argv)/* TODO: Relax this once comparison is byte-based, and we no longer * have to worry about sector alignment */ assert (QEMU_IS_ALIGNED (pnum1 | pnum2,BDRV_SECTOR_SIZE )); -if (pnum1) { - nb_sectors =MIN (nb_sectors, pnum1 >>BDRV_SECTOR_BITS ); - } -if (pnum2) { - nb_sectors =MIN (nb_sectors, pnum2 >>BDRV_SECTOR_BITS ); - } + +assert (pnum1 && pnum2); + nb_sectors =MIN (pnum1, pnum2) >>BDRV_SECTOR_BITS ;if (strict) {if (status1 != status2) {
@@ -1423,9 +1411,10 @@ static int img_compare(int argc, char **argv) } }if ((status1 &BDRV_BLOCK_ZERO ) && (status2 &BDRV_BLOCK_ZERO )) { - nb_sectors =DIV_ROUND_UP (MIN (pnum1, pnum2),BDRV_SECTOR_SIZE ); +/* nothing to do */ }else if (allocated1 == allocated2) {if (allocated1) { + nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); ret =blk_pread (blk1, sector_num <<BDRV_SECTOR_BITS , buf1, nb_sectors <<BDRV_SECTOR_BITS );if (ret <0 ) {
@@ -1454,7 +1443,7 @@ static int img_compare(int argc, char **argv) } } }else { - + nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS );if (allocated1) { ret =check_empty_sectors (blk1, sector_num, nb_sectors, filename1, buf1, quiet);
@@ -1477,30 +1466,24 @@ static int img_compare(int argc, char **argv)if (total_sectors1 != total_sectors2) {BlockBackend *blk_over; -int64_t total_sectors_over;const char *filename_over;qprintf (quiet,"Warning: Image size mismatch!\n" );if (total_sectors1 > total_sectors2) { - total_sectors_over = total_sectors1; blk_over = blk1; filename_over = filename1; }else { - total_sectors_over = total_sectors2; blk_over = blk2; filename_over = filename2; } -for (;;) { +while (sector_num < progress_base) {int64_t count; - nb_sectors =sectors_to_process (total_sectors_over, sector_num); -if (nb_sectors <=0 ) { -break ; - } ret =bdrv_is_allocated_above (blk_bs (blk_over),NULL , sector_num *BDRV_SECTOR_SIZE , - nb_sectors *BDRV_SECTOR_SIZE , + (progress_base - sector_num) * +BDRV_SECTOR_SIZE , &count);if (ret <0 ) { ret =3 ;
@@ -1514,6 +1497,7 @@ static int img_compare(int argc, char **argv)assert (QEMU_IS_ALIGNED (count,BDRV_SECTOR_SIZE )); nb_sectors = count >>BDRV_SECTOR_BITS ;if (ret) { + nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); ret =check_empty_sectors (blk_over, sector_num, nb_sectors, filename_over, buf1, quiet);if (ret) {
Signature · 1 line
2.13.6
PATCH v6 14/24 qemu-img: Speed up compare on pre-allocated larger file
In reply to Eric BlakeCompare the following images with all-zero contents:
$ truncate --size 1M A $ qemu-img create -f qcow2 -o preallocation=off B 1G $ qemu-img create -f qcow2 -o preallocation=metadata C 1G
On my machine, the difference is noticeable for pre-patch speeds, with more than an order of magnitude in difference caused by the choice of preallocation in the qcow2 file:
$ time ./qemu-img compare -f raw -F qcow2 A B Warning: Image size mismatch! Images are identical.
real 0m0.014s
user 0m0.007s
sys 0m0.007s
$ time ./qemu-img compare -f raw -F qcow2 A C Warning: Image size mismatch! Images are identical.
real 0m0.341s
user 0m0.144s
sys 0m0.188s
Why? Because bdrv_is_allocated() returns false for image B but true for image C, throwing away the fact that both images know via lseek(SEEK_HOLE) that the entire image still reads as zero.
Quoted 1 line
From there, qemu-img ends up calling
bdrv_pread()for every byte
of the tail, instead of quickly looking for the next allocation. The solution: use block_status instead of is_allocated, giving:
$ time ./qemu-img compare -f raw -F qcow2 A C Warning: Image size mismatch! Images are identical.
real 0m0.014s
user 0m0.011s
sys 0m0.003s
which is on par with the speeds for no pre-allocation.
---
v6: rebase to interface change [Kevin], minor enough to keep R-b
v4-v5: no change
v3: new patch
--- qemu-img.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
qemu-img.c +4 -4
@@ -1480,11 +1480,11 @@ static int img_compare(int argc, char **argv)while (sector_num < progress_base) {int64_t count; - ret =bdrv_is_allocated_above (blk_bs (blk_over),NULL , + ret =bdrv_block_status_above (blk_bs (blk_over),NULL , sector_num *BDRV_SECTOR_SIZE , (progress_base - sector_num) *BDRV_SECTOR_SIZE , - &count); + &count,NULL ,NULL );if (ret <0 ) { ret =3 ; error_report("Sector allocation test failed for %s",
@@ -1492,11 +1492,11 @@ static int img_compare(int argc, char **argv) goto out; } -/* TODO relax this once bdrv_is_allocated_above does not enforce +/* TODO relax this once bdrv_block_status_above does not enforce * sector alignment */ assert (QEMU_IS_ALIGNED (count,BDRV_SECTOR_SIZE )); nb_sectors = count >>BDRV_SECTOR_BITS ; -if (ret) { +if (ret &BDRV_BLOCK_ALLOCATED && !(ret &BDRV_BLOCK_ZERO )) { nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); ret =check_empty_sectors (blk_over, sector_num, nb_sectors, filename_over, buf1, quiet);
Signature · 1 line
2.13.6
PATCH v6 15/24 qemu-img: Add find_nonzero()
In reply to Eric BlakeDuring 'qemu-img compare', when we are checking that an allocated portion of one file is all zeros, we don't need to waste time computing how many additional sectors after the first non-zero byte are also non-zero. Create a new helper find_nonzero() to do the check for a first non-zero sector, and rebase check_empty_sectors() to use it.
The new interface intentionally uses bytes in its interface, even though it still crawls the buffer a sector at a time; it is robust to a partial sector at the end of the buffer.
---
v4-v5: no change
v3: new patch
--- qemu-img.c | 32 ++++++++++++++++++++++++++++---- 1 file changed, 28 insertions(+), 4 deletions(-)
qemu-img.c +28 -4
@@ -1065,6 +1065,28 @@ done: }/* +* Returns -1 if 'buf' contains only zeroes, otherwise the byte index +* of the first sector boundary within buf where the sector contains a +* non-zero byte. This function is robust to a buffer that is not +* sector-aligned. +*/ +static int64_t find_nonzero (const uint8_t *buf,int64_t n) +{ +int64_t i; +int64_t end =QEMU_ALIGN_DOWN (n,BDRV_SECTOR_SIZE ); + +for (i =0 ; i < end; i +=BDRV_SECTOR_SIZE ) { +if (!buffer_is_zero (buf + i,BDRV_SECTOR_SIZE )) { +return i; + } + } +if (i < n && !buffer_is_zero (buf + i, n - end)) { +return i; + } +return -1 ; +} + +/* * Returns true iff the first sector pointed to by 'buf' contains at least * a non-NUL byte. *
@@ -1189,7 +1211,9 @@ static int check_empty_sectors(BlockBackend *blk, int64_t sect_num, int sect_count, const char *filename, uint8_t *buffer, bool quiet) { - int pnum, ret = 0; + int ret = 0; +int64_t idx; + ret =blk_pread (blk, sect_num <<BDRV_SECTOR_BITS , buffer, sect_count <<BDRV_SECTOR_BITS );if (ret <0 ) {
@@ -1197,10 +1221,10 @@ static int check_empty_sectors(BlockBackend *blk, int64_t sect_num, sectors_to_bytes(sect_num), filename, strerror(-ret));return ret; } - ret =is_allocated_sectors (buffer, sect_count, &pnum); -if (ret || pnum != sect_count) { + idx =find_nonzero (buffer, sect_count *BDRV_SECTOR_SIZE ); +if (idx >=0 ) {qprintf (quiet,"Content mismatch at offset %" PRId64"!\n" , -sectors_to_bytes (ret ? sect_num : sect_num + pnum)); +sectors_to_bytes (sect_num) + idx);return 1 ; }
Signature · 1 line
2.13.6
PATCH v6 16/24 qemu-img: Drop redundant error message in compare
In reply to Eric BlakeIf a read error is encountered during 'qemu-img compare', we were printing the "Error while reading offset ..." message twice; this was because our helper function was awkward, printing output on some but not all paths. Fix it to consistently report errors on all paths, so that the callers do not risk a redundant message, and update the testsuite for the improved output.
Further simplify the code by hoisting the conversion from an error message to an exit code into the helper function, rather than repeating that logic at all callers (yes, the helper function is now less generic, but it's a net win in lines of code).
---
v5: tweak commit message, but no code change
v4: no change
v3: new patch
--- qemu-img.c | 19 +++++-------------- tests/qemu-iotests/074.out | 2 -- 2 files changed, 5 insertions(+), 16 deletions(-)
qemu-img.c +5 -14
@@ -1197,8 +1197,10 @@ static int64_t sectors_to_bytes(int64_t sectors)/* * Check if passed sectors are empty (not allocated or contain only 0 bytes) * -* Returns 0 in case sectors are filled with 0, 1 if sectors contain non-zero -* data and negative value on error. +* Intended for use by 'qemu-img compare': Returns 0 in case sectors are +* filled with 0, 1 if sectors contain non-zero data (this is a comparison +* failure), and 4 on error (the exit status for read errors), after emitting +* an error message. * * @param blk: BlockBackend for the image * @param sect_num: Number of first sector to check
@@ -1219,7 +1221,7 @@ static int check_empty_sectors(BlockBackend *blk, int64_t sect_num,if (ret <0 ) {error_report ("Error while reading offset %" PRId64" of %s: %s" ,sectors_to_bytes (sect_num), filename,strerror (-ret)); -return ret; +return 4 ; } idx =find_nonzero (buffer, sect_count *BDRV_SECTOR_SIZE );if (idx >=0 ) {
@@ -1476,11 +1478,6 @@ static int img_compare(int argc, char **argv) filename2, buf1, quiet); }if (ret) { -if (ret <0 ) { -error_report ("Error while reading offset %" PRId64": %s" , -sectors_to_bytes (sector_num),strerror (-ret)); - ret =4 ; - } goto out; } }
@@ -1525,12 +1522,6 @@ static int img_compare(int argc, char **argv) ret =check_empty_sectors (blk_over, sector_num, nb_sectors, filename_over, buf1, quiet);if (ret) { -if (ret <0 ) { -error_report ("Error while reading offset %" PRId64 -" of %s: %s" ,sectors_to_bytes (sector_num), - filename_over,strerror (-ret)); - ret =4 ; - } goto out; } }
tests/qemu-iotests/074.out +0 -2
@@ -4,7 +4,6 @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1073741824 wrote 512/512 bytes at offset 512 512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) qemu-img: Error while reading offset 0 of blkdebug:TEST_DIR/blkdebug.conf:TEST_DIR/t.IMGFMT: Input/output error -qemu-img: Error while reading offset 0: Input/output error 4 Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1073741824 Formatting 'TEST_DIR/t.IMGFMT.2', fmt=IMGFMT size=0
@@ -12,7 +11,6 @@ Formatting 'TEST_DIR/t.IMGFMT.2', fmt=IMGFMT size=0 wrote 512/512 bytes at offset 512 512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) qemu-img: Error while reading offset 0 of blkdebug:TEST_DIR/blkdebug.conf:TEST_DIR/t.IMGFMT: Input/output error -qemu-img: Error while reading offset 0 of blkdebug:TEST_DIR/blkdebug.conf:TEST_DIR/t.IMGFMT: Input/output error Warning: Image size mismatch! 4 Cleanup
Signature · 1 line
2.13.6
PATCH v6 17/24 qemu-img: Change check_empty_sectors() to byte-based
In reply to Eric BlakeContinue on the quest to make more things byte-based instead of sector-based.
---
v4-v5: no change
v3: new patch
--- qemu-img.c | 27 +++++++++++++++------------ 1 file changed, 15 insertions(+), 12 deletions(-)
qemu-img.c +15 -12
@@ -1203,30 +1203,29 @@ static int64_t sectors_to_bytes(int64_t sectors)* an error message. * * @param blk: BlockBackend for the image -* @param sect_num: Number of first sector to check -* @param sect_count: Number of sectors to check +* @param offset: Starting offset to check +* @param bytes: Number of bytes to check * @param filename: Name of disk file we are checking (logging purpose) * @param buffer: Allocated buffer for storing read data * @param quiet: Flag for quiet mode */ -static int check_empty_sectors (BlockBackend *blk,int64_t sect_num, -int sect_count,const char *filename, +static int check_empty_sectors (BlockBackend *blk,int64_t offset, +int64_t bytes,const char *filename,uint8_t *buffer,bool quiet) {int ret =0 ;int64_t idx; - ret =blk_pread (blk, sect_num <<BDRV_SECTOR_BITS , buffer, - sect_count <<BDRV_SECTOR_BITS ); + ret =blk_pread (blk, offset, buffer, bytes);if (ret <0 ) {error_report ("Error while reading offset %" PRId64" of %s: %s" , -sectors_to_bytes (sect_num), filename,strerror (-ret)); + offset, filename,strerror (-ret));return 4 ; } - idx =find_nonzero (buffer, sect_count *BDRV_SECTOR_SIZE ); + idx =find_nonzero (buffer, bytes);if (idx >=0 ) {qprintf (quiet,"Content mismatch at offset %" PRId64"!\n" , -sectors_to_bytes (sect_num) + idx); + offset + idx);return 1 ; }
@@ -1471,10 +1470,12 @@ static int img_compare(int argc, char **argv) }else { nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS );if (allocated1) { - ret =check_empty_sectors (blk1, sector_num, nb_sectors, + ret =check_empty_sectors (blk1, sector_num *BDRV_SECTOR_SIZE , + nb_sectors *BDRV_SECTOR_SIZE , filename1, buf1, quiet); }else { - ret =check_empty_sectors (blk2, sector_num, nb_sectors, + ret =check_empty_sectors (blk2, sector_num *BDRV_SECTOR_SIZE , + nb_sectors *BDRV_SECTOR_SIZE , filename2, buf1, quiet); }if (ret) {
@@ -1519,7 +1520,9 @@ static int img_compare(int argc, char **argv) nb_sectors = count >>BDRV_SECTOR_BITS ;if (ret &BDRV_BLOCK_ALLOCATED && !(ret &BDRV_BLOCK_ZERO )) { nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); - ret =check_empty_sectors (blk_over, sector_num, nb_sectors, + ret =check_empty_sectors (blk_over, + sector_num *BDRV_SECTOR_SIZE , + nb_sectors *BDRV_SECTOR_SIZE , filename_over, buf1, quiet);if (ret) { goto out;
Signature · 1 line
2.13.6
PATCH v6 18/24 qemu-img: Change compare_sectors() to be byte-based
In reply to Eric BlakeIn the continuing quest to make more things byte-based, change compare_sectors(), renaming it to compare_buffers() in the process. Note that one caller (qemu-img compare) only cares about the first difference, while the other (qemu-img rebase) cares about how many consecutive sectors have the same equal/different status; however, this patch does not bother to micro-optimize the compare case to avoid the comparisons of sectors beyond the first mismatch. Both callers are always passing valid buffers in, so the initial check for buffer size can be turned into an assertion.
---
v4-v5: no change
v3: new patch
--- qemu-img.c | 55 +++++++++++++++++++++++++++---------------------------- 1 file changed, 27 insertions(+), 28 deletions(-)
qemu-img.c +27 -28
@@ -1156,31 +1156,28 @@ static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum, }/* -* Compares two buffers sector by sector. Returns 0 if the first sector of both -* buffers matches, non-zero otherwise. +* Compares two buffers sector by sector. Returns 0 if the first +* sector of each buffer matches, non-zero otherwise. * -* pnum is set to the number of sectors (including and immediately following -* the first one) that are known to have the same comparison result +* pnum is set to the sector-aligned size of the buffer prefix that +* has the same matching status as the first sector. */ -static int compare_sectors (const uint8_t *buf1,const uint8_t *buf2,int n, -int *pnum) +static int compare_buffers (const uint8_t *buf1,const uint8_t *buf2, +int64_t bytes,int64_t *pnum) {bool res; -int i; +int64_t i =MIN (bytes,BDRV_SECTOR_SIZE ); -if (n <=0 ) { - *pnum =0 ; -return 0 ; - } +assert (bytes >0 ); - res = !!memcmp (buf1, buf2,512 ); -for (i =1 ; i < n; i++) { - buf1 +=512 ; - buf2 +=512 ; + res = !!memcmp (buf1, buf2, i); +while (i < bytes) { +int64_t len =MIN (bytes - i,BDRV_SECTOR_SIZE ); -if (!!memcmp (buf1, buf2,512 ) != res) { +if (!!memcmp (buf1 + i, buf2 + i, len) != res) {break ; } + i += len; } *pnum = i;
@@ -1255,7 +1252,7 @@ static int img_compare(int argc, char **argv)int64_t total_sectors;int64_t sector_num =0 ;int64_t nb_sectors; -int c, pnum; +int c;uint64_t progress_base;bool image_opts = false;bool force_share = false;
@@ -1439,6 +1436,8 @@ static int img_compare(int argc, char **argv)/* nothing to do */ }else if (allocated1 == allocated2) {if (allocated1) { +int64_t pnum; + nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); ret =blk_pread (blk1, sector_num <<BDRV_SECTOR_BITS , buf1, nb_sectors <<BDRV_SECTOR_BITS );
@@ -1458,11 +1457,11 @@ static int img_compare(int argc, char **argv) ret =4 ; goto out; } - ret =compare_sectors (buf1, buf2, nb_sectors, &pnum); -if (ret || pnum != nb_sectors) { + ret =compare_buffers (buf1, buf2, + nb_sectors *BDRV_SECTOR_SIZE , &pnum); +if (ret || pnum != nb_sectors *BDRV_SECTOR_SIZE ) {qprintf (quiet,"Content mismatch at offset %" PRId64"!\n" , -sectors_to_bytes ( - ret ? sector_num : sector_num + pnum)); +sectors_to_bytes (sector_num) + (ret ?0 : pnum)); ret =1 ; goto out; }
@@ -3354,16 +3353,16 @@ static int img_rebase(int argc, char **argv)/* If they differ, we need to write to the COW file */ uint64_t written =0 ; -while (written < n) { -int pnum; +while (written < n *BDRV_SECTOR_SIZE ) { +int64_t pnum; -if (compare_sectors (buf_old + written *512 , - buf_new + written *512 , n - written, &pnum)) +if (compare_buffers (buf_old + written, + buf_new + written, + n *BDRV_SECTOR_SIZE - written, &pnum)) { ret =blk_pwrite (blk, - (sector + written) <<BDRV_SECTOR_BITS , - buf_old + written *512 , - pnum <<BDRV_SECTOR_BITS ,0 ); + (sector <<BDRV_SECTOR_BITS ) + written, + buf_old + written, pnum,0 );if (ret <0 ) {error_report ("Error while writing to COW image: %s" ,strerror (-ret));
Signature · 1 line
2.13.6
PATCH v6 19/24 qemu-img: Change img_rebase() to be byte-based
In reply to Eric BlakeIn the continuing quest to make more things byte-based, change the internal iteration of img_rebase(). We can finally drop the TODO assertion added earlier, now that the entire algorithm is byte-based and no longer has to shift from bytes to sectors.
Most of the change is mechanical ('num_sectors' becomes 'size', 'sector' becomes 'offset', 'n' goes from sectors to bytes); some of it is also a cleanup (use of MIN() instead of open-coding, loss of variable 'count' added earlier in commit d6a644bb).
---
v4-v5: no change
v3: new patch
--- qemu-img.c | 84 +++++++++++++++++++++++++------------------------------------- 1 file changed, 34 insertions(+), 50 deletions(-)
qemu-img.c +34 -50
@@ -3248,70 +3248,58 @@ static int img_rebase(int argc, char **argv)* the image is the same as the original one at any time. */ if (!unsafe) { -int64_t num_sectors; -int64_t old_backing_num_sectors; -int64_t new_backing_num_sectors =0 ; -uint64_t sector; -int n; -int64_t count; +int64_t size; +int64_t old_backing_size; +int64_t new_backing_size =0 ; +uint64_t offset; +int64_t n;float local_progress =0 ; buf_old =blk_blockalign (blk,IO_BUF_SIZE ); buf_new =blk_blockalign (blk,IO_BUF_SIZE ); - num_sectors =blk_nb_sectors (blk); -if (num_sectors <0 ) { + size =blk_getlength (blk); +if (size <0 ) {error_report ("Could not get size of '%s': %s" , - filename,strerror (-num_sectors)); + filename,strerror (-size)); ret =-1 ; goto out; } - old_backing_num_sectors =blk_nb_sectors (blk_old_backing); -if (old_backing_num_sectors <0 ) { + old_backing_size =blk_getlength (blk_old_backing); +if (old_backing_size <0 ) {char backing_name[PATH_MAX ];bdrv_get_backing_filename (bs, backing_name,sizeof (backing_name));error_report ("Could not get size of '%s': %s" , - backing_name,strerror (-old_backing_num_sectors)); + backing_name,strerror (-old_backing_size)); ret =-1 ; goto out; }if (blk_new_backing) { - new_backing_num_sectors =blk_nb_sectors (blk_new_backing); -if (new_backing_num_sectors <0 ) { + new_backing_size =blk_getlength (blk_new_backing); +if (new_backing_size <0 ) {error_report ("Could not get size of '%s': %s" , - out_baseimg,strerror (-new_backing_num_sectors)); + out_baseimg,strerror (-new_backing_size)); ret =-1 ; goto out; } } -if (num_sectors !=0 ) { - local_progress = (float )100 / - (num_sectors /MIN (num_sectors,IO_BUF_SIZE /512 )); +if (size !=0 ) { + local_progress = (float )100 / (size /MIN (size,IO_BUF_SIZE )); } -for (sector =0 ; sector < num_sectors; sector += n) { - -/* How many sectors can we handle with the next read? */ -if (sector + (IO_BUF_SIZE /512 ) <= num_sectors) { - n = (IO_BUF_SIZE /512 ); - }else { - n = num_sectors - sector; - } +for (offset =0 ; offset < size; offset += n) { +/* How many bytes can we handle with the next read? */ + n =MIN (IO_BUF_SIZE , size - offset);/* If the cluster is allocated, we don't need to take action */ - ret =bdrv_is_allocated (bs, sector <<BDRV_SECTOR_BITS , - n <<BDRV_SECTOR_BITS , &count); + ret =bdrv_is_allocated (bs, offset, n, &n);if (ret <0 ) {error_report ("error while reading image metadata: %s" ,strerror (-ret)); goto out; } -/* TODO relax this once bdrv_is_allocated does not enforce -* sector alignment */ -assert (QEMU_IS_ALIGNED (count,BDRV_SECTOR_SIZE )); - n = count >>BDRV_SECTOR_BITS ;if (ret) {continue ; }
@@ -3320,30 +3308,28 @@ static int img_rebase(int argc, char **argv)* Read old and new backing file and take into consideration that * backing files may be smaller than the COW image. */ -if (sector >= old_backing_num_sectors) { -memset (buf_old,0 , n *BDRV_SECTOR_SIZE ); +if (offset >= old_backing_size) { +memset (buf_old,0 , n); }else { -if (sector + n > old_backing_num_sectors) { - n = old_backing_num_sectors - sector; +if (offset + n > old_backing_size) { + n = old_backing_size - offset; } - ret =blk_pread (blk_old_backing, sector <<BDRV_SECTOR_BITS , - buf_old, n <<BDRV_SECTOR_BITS ); + ret =blk_pread (blk_old_backing, offset, buf_old, n);if (ret <0 ) {error_report ("error while reading from old backing file" ); goto out; } } -if (sector >= new_backing_num_sectors || !blk_new_backing) { -memset (buf_new,0 , n *BDRV_SECTOR_SIZE ); +if (offset >= new_backing_size || !blk_new_backing) { +memset (buf_new,0 , n); }else { -if (sector + n > new_backing_num_sectors) { - n = new_backing_num_sectors - sector; +if (offset + n > new_backing_size) { + n = new_backing_size - offset; } - ret =blk_pread (blk_new_backing, sector <<BDRV_SECTOR_BITS , - buf_new, n <<BDRV_SECTOR_BITS ); + ret =blk_pread (blk_new_backing, offset, buf_new, n);if (ret <0 ) {error_report ("error while reading from new backing file" ); goto out;
@@ -3353,15 +3339,13 @@ static int img_rebase(int argc, char **argv)/* If they differ, we need to write to the COW file */ uint64_t written =0 ; -while (written < n *BDRV_SECTOR_SIZE ) { +while (written < n) {int64_t pnum; -if (compare_buffers (buf_old + written, - buf_new + written, - n *BDRV_SECTOR_SIZE - written, &pnum)) +if (compare_buffers (buf_old + written, buf_new + written, + n - written, &pnum)) { - ret =blk_pwrite (blk, - (sector <<BDRV_SECTOR_BITS ) + written, + ret =blk_pwrite (blk, offset + written, buf_old + written, pnum,0 );if (ret <0 ) { error_report("Error while writing to COW image: %s",
Signature · 1 line
2.13.6
PATCH v6 20/24 qemu-img: Change img_compare() to be byte-based
In reply to Eric BlakeIn the continuing quest to make more things byte-based, change the internal iteration of img_compare(). We can finally drop the TODO assertions added earlier, now that the entire algorithm is byte-based and no longer has to shift from bytes to sectors.
Most of the change is mechanical ('total_sectors' becomes 'total_size', 'sector_num' becomes 'offset', 'nb_sectors' becomes 'chunk', 'progress_base' goes from sectors to bytes); some of it is also a cleanup (sectors_to_bytes() is now unused, loss of variable 'count' added earlier in commit 51b0a488).
---
v6: rebase to split return interface change [Kevin], minor enough to keep R-b
v5: rebase to earlier change, minor enough to keep R-b
v4: no change
v3: new patch
--- qemu-img.c | 124 ++++++++++++++++++++++++------------------------------------- 1 file changed, 48 insertions(+), 76 deletions(-)
qemu-img.c +48 -76
@@ -1186,11 +1186,6 @@ static int compare_buffers(const uint8_t *buf1, const uint8_t *buf2,#define IO_BUF_SIZE (2 * 1024 * 1024) -static int64_t sectors_to_bytes (int64_t sectors) -{ -return sectors <<BDRV_SECTOR_BITS ; -} -/* * Check if passed sectors are empty (not allocated or contain only 0 bytes) *
@@ -1241,7 +1236,7 @@ static int img_compare(int argc, char **argv)const char *fmt1 =NULL , *fmt2 =NULL , *cache, *filename1, *filename2;BlockBackend *blk1, *blk2;BlockDriverState *bs1, *bs2; -int64_t total_sectors1, total_sectors2; +int64_t total_size1, total_size2;uint8_t *buf1 =NULL , *buf2 =NULL ;int64_t pnum1, pnum2;int allocated1, allocated2;
@@ -1249,9 +1244,9 @@ static int img_compare(int argc, char **argv)bool progress = false, quiet = false, strict = false;int flags;bool writethrough; -int64_t total_sectors; -int64_t sector_num =0 ; -int64_t nb_sectors; +int64_t total_size; +int64_t offset =0 ; +int64_t chunk;int c;uint64_t progress_base;bool image_opts = false;
@@ -1365,39 +1360,37 @@ static int img_compare(int argc, char **argv) buf1 =blk_blockalign (blk1,IO_BUF_SIZE ); buf2 =blk_blockalign (blk2,IO_BUF_SIZE ); - total_sectors1 =blk_nb_sectors (blk1); -if (total_sectors1 <0 ) { + total_size1 =blk_getlength (blk1); +if (total_size1 <0 ) {error_report ("Can't get size of %s: %s" , - filename1,strerror (-total_sectors1)); + filename1,strerror (-total_size1)); ret =4 ; goto out; } - total_sectors2 =blk_nb_sectors (blk2); -if (total_sectors2 <0 ) { + total_size2 =blk_getlength (blk2); +if (total_size2 <0 ) {error_report ("Can't get size of %s: %s" , - filename2,strerror (-total_sectors2)); + filename2,strerror (-total_size2)); ret =4 ; goto out; } - total_sectors =MIN (total_sectors1, total_sectors2); - progress_base =MAX (total_sectors1, total_sectors2); + total_size =MIN (total_size1, total_size2); + progress_base =MAX (total_size1, total_size2);qemu_progress_print (0 ,100 ); -if (strict && total_sectors1 != total_sectors2) { +if (strict && total_size1 != total_size2) { ret =1 ;qprintf (quiet,"Strict mode: Image size mismatch!\n" ); goto out; } -while (sector_num < total_sectors) { +while (offset < total_size) {int status1, status2; - status1 =bdrv_block_status_above (bs1,NULL , - sector_num *BDRV_SECTOR_SIZE , - (total_sectors1 - sector_num) * -BDRV_SECTOR_SIZE , - &pnum1,NULL ,NULL ); + status1 =bdrv_block_status_above (bs1,NULL , offset, + total_size1 - offset, &pnum1,NULL , +NULL );if (status1 <0 ) { ret =3 ;error_report ("Sector allocation test failed for %s" , filename1);
@@ -1405,30 +1398,24 @@ static int img_compare(int argc, char **argv) } allocated1 = status1 &BDRV_BLOCK_ALLOCATED ; - status2 =bdrv_block_status_above (bs2,NULL , - sector_num *BDRV_SECTOR_SIZE , - (total_sectors2 - sector_num) * -BDRV_SECTOR_SIZE , - &pnum2,NULL ,NULL ); + status2 =bdrv_block_status_above (bs2,NULL , offset, + total_size2 - offset, &pnum2,NULL , +NULL );if (status2 <0 ) { ret =3 ;error_report ("Sector allocation test failed for %s" , filename2); goto out; } allocated2 = status2 &BDRV_BLOCK_ALLOCATED ; -/* TODO: Relax this once comparison is byte-based, and we no longer -* have to worry about sector alignment */ -assert (QEMU_IS_ALIGNED (pnum1 | pnum2,BDRV_SECTOR_SIZE ));assert (pnum1 && pnum2); - nb_sectors =MIN (pnum1, pnum2) >>BDRV_SECTOR_BITS ; + chunk =MIN (pnum1, pnum2);if (strict) {if (status1 != status2) { ret =1 ;qprintf (quiet,"Strict mode: Offset %" PRId64 -" block status mismatch!\n" , -sectors_to_bytes (sector_num)); +" block status mismatch!\n" , offset); goto out; } }
@@ -1438,59 +1425,54 @@ static int img_compare(int argc, char **argv)if (allocated1) {int64_t pnum; - nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); - ret =blk_pread (blk1, sector_num <<BDRV_SECTOR_BITS , buf1, - nb_sectors <<BDRV_SECTOR_BITS ); + chunk =MIN (chunk,IO_BUF_SIZE ); + ret =blk_pread (blk1, offset, buf1, chunk);if (ret <0 ) { -error_report ("Error while reading offset %" PRId64" of %s:" -" %s" ,sectors_to_bytes (sector_num), filename1, -strerror (-ret)); +error_report ("Error while reading offset %" PRId64 +" of %s: %s" , + offset, filename1,strerror (-ret)); ret =4 ; goto out; } - ret =blk_pread (blk2, sector_num <<BDRV_SECTOR_BITS , buf2, - nb_sectors <<BDRV_SECTOR_BITS ); + ret =blk_pread (blk2, offset, buf2, chunk);if (ret <0 ) {error_report ("Error while reading offset %" PRId64 -" of %s: %s" ,sectors_to_bytes (sector_num), - filename2,strerror (-ret)); +" of %s: %s" , + offset, filename2,strerror (-ret)); ret =4 ; goto out; } - ret =compare_buffers (buf1, buf2, - nb_sectors *BDRV_SECTOR_SIZE , &pnum); -if (ret || pnum != nb_sectors *BDRV_SECTOR_SIZE ) { + ret =compare_buffers (buf1, buf2, chunk, &pnum); +if (ret || pnum != chunk) {qprintf (quiet,"Content mismatch at offset %" PRId64"!\n" , -sectors_to_bytes (sector_num) + (ret ?0 : pnum)); + offset + (ret ?0 : pnum)); ret =1 ; goto out; } } }else { - nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); + chunk =MIN (chunk,IO_BUF_SIZE );if (allocated1) { - ret =check_empty_sectors (blk1, sector_num *BDRV_SECTOR_SIZE , - nb_sectors *BDRV_SECTOR_SIZE , + ret =check_empty_sectors (blk1, offset, chunk, filename1, buf1, quiet); }else { - ret =check_empty_sectors (blk2, sector_num *BDRV_SECTOR_SIZE , - nb_sectors *BDRV_SECTOR_SIZE , + ret =check_empty_sectors (blk2, offset, chunk, filename2, buf1, quiet); }if (ret) { goto out; } } - sector_num += nb_sectors; -qemu_progress_print (((float ) nb_sectors / progress_base)*100 ,100 ); + offset += chunk; +qemu_progress_print (((float ) chunk / progress_base) *100 ,100 ); } -if (total_sectors1 != total_sectors2) { +if (total_size1 != total_size2) {BlockBackend *blk_over;const char *filename_over;qprintf (quiet,"Warning: Image size mismatch!\n" ); -if (total_sectors1 > total_sectors2) { +if (total_size1 > total_size2) { blk_over = blk1; filename_over = filename1; }else {
@@ -1498,14 +1480,10 @@ static int img_compare(int argc, char **argv) filename_over = filename2; } -while (sector_num < progress_base) { -int64_t count; - - ret =bdrv_block_status_above (blk_bs (blk_over),NULL , - sector_num *BDRV_SECTOR_SIZE , - (progress_base - sector_num) * -BDRV_SECTOR_SIZE , - &count,NULL ,NULL ); +while (offset < progress_base) { + ret =bdrv_block_status_above (blk_bs (blk_over),NULL , offset, + progress_base - offset, &chunk, +NULL ,NULL );if (ret <0 ) { ret =3 ; error_report("Sector allocation test failed for %s",
@@ -1513,22 +1491,16 @@ static int img_compare(int argc, char **argv) goto out; } -/* TODO relax this once bdrv_block_status_above does not enforce -* sector alignment */ -assert (QEMU_IS_ALIGNED (count,BDRV_SECTOR_SIZE )); - nb_sectors = count >>BDRV_SECTOR_BITS ;if (ret &BDRV_BLOCK_ALLOCATED && !(ret &BDRV_BLOCK_ZERO )) { - nb_sectors =MIN (nb_sectors,IO_BUF_SIZE >>BDRV_SECTOR_BITS ); - ret =check_empty_sectors (blk_over, - sector_num *BDRV_SECTOR_SIZE , - nb_sectors *BDRV_SECTOR_SIZE , + chunk =MIN (chunk,IO_BUF_SIZE ); + ret =check_empty_sectors (blk_over, offset, chunk, filename_over, buf1, quiet);if (ret) { goto out; } } - sector_num += nb_sectors; -qemu_progress_print (((float ) nb_sectors / progress_base)*100 ,100 ); + offset += chunk; +qemu_progress_print (((float ) chunk / progress_base) *100 ,100 ); } }
Signature · 1 line
2.13.6
PATCH v6 22/24 block: Relax bdrv_aligned_preadv() assertion
In reply to Eric BlakeNow that bdrv_is_allocated accepts non-aligned inputs, we can remove the TODO added in commit d6a644bb.
---
v4-v5: no change
v3: new patch [Kevin]
--- block/io.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
block/io.c +2 -6
@@ -1104,18 +1104,14 @@ static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child, }if (flags &BDRV_REQ_COPY_ON_READ ) { -/* TODO: Simplify further once bdrv_is_allocated no longer -* requires sector alignment */ -int64_t start =QEMU_ALIGN_DOWN (offset,BDRV_SECTOR_SIZE ); -int64_t end =QEMU_ALIGN_UP (offset + bytes,BDRV_SECTOR_SIZE );int64_t pnum; - ret =bdrv_is_allocated (bs, start, end - start, &pnum); + ret =bdrv_is_allocated (bs, offset, bytes, &pnum);if (ret <0 ) { goto out; } -if (!ret || pnum != end - start) { +if (!ret || pnum != bytes) { ret =bdrv_co_do_copy_on_readv (child, offset, bytes, qiov); goto out; }
Signature · 1 line
2.13.6
PATCH v6 21/24 block: Align block status requests
In reply to Eric BlakeAny device that has request_alignment greater than 512 should be unable to report status at a finer granularity; it may also be simpler for such devices to be guaranteed that the block layer has rounded things out to the granularity boundary (the way the block layer already rounds all other I/O out). Besides, getting the code correct for super-sector alignment also benefits us for the fact that our public interface now has byte granularity, even though none of our drivers have byte-level callbacks.
Add an assertion in blkdebug that proves that the block layer never requests status of unaligned sections, similar to what it does on other requests (while still keeping the generic helper in place for when future patches add a throttle driver). Note that iotest 177 already covers this (it would fail if you use just the blkdebug.c hunk without the io.c changes). Meanwhile, we can drop assertions in callers that no longer have to pass in sector-aligned addresses.
There is a mid-function scope added for 'count' and 'longret', for a couple of reasons: first, an upcoming patch will add an 'if' statement that checks whether a driver has an old- or new-style callback, and can conveniently use the same scope for less indentation churn at that time. Second, since we are trying to get rid of sector-based computations, wrapping things in a scope makes it easier to group and see what will be deleted in a final cleanup patch once all drivers have been converted to the new-style callback.
---
v6: rebase to split return interface change [Kevin]
v5: rebase to earlier changes, add more comments
v4: no change
v3: tweak commit message [Fam], rebase to context conflicts, ensure we don't exceed 32-bit limit, drop R-b
v2: new patch
--- include/block/block_int.h | 3 +- block/io.c | 71 ++++++++++++++++++++++++++++++----------------- block/blkdebug.c | 13 ++++++++- 3 files changed, 59 insertions(+), 28 deletions(-)
include/block/block_int.h +2 -1
@@ -207,7 +207,8 @@ struct BlockDriver {* according to the current layer, and should not set * BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW. See block.h * for the meaning of _DATA, _ZERO, and _OFFSET_VALID. The block -* layer guarantees non-NULL pnum and file. +* layer guarantees input aligned to request_alignment, as well as +* non-NULL pnum and file. */ int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
block/io.c +45 -26
101 lines not on this page · raw message
block/blkdebug.c +12 -1
@@ -627,6 +627,17 @@ static int coroutine_fn blkdebug_co_pdiscard(BlockDriverState *bs,return bdrv_co_pdiscard (bs->file->bs, offset, bytes); } +static int64_t coroutine_fn blkdebug_co_get_block_status( +BlockDriverState *bs,int64_t sector_num,int nb_sectors,int *pnum, +BlockDriverState **file) +{ +assert (QEMU_IS_ALIGNED (sector_num | nb_sectors, +DIV_ROUND_UP (bs->bl.request_alignment, +BDRV_SECTOR_SIZE ))); +return bdrv_co_get_block_status_from_file (bs, sector_num, nb_sectors, + pnum, file); +} +static void blkdebug_close (BlockDriverState *bs) {BDRVBlkdebugState *s = bs->opaque;
@@ -896,7 +907,7 @@ static BlockDriver bdrv_blkdebug = { .bdrv_co_flush_to_disk = blkdebug_co_flush, .bdrv_co_pwrite_zeroes = blkdebug_co_pwrite_zeroes, .bdrv_co_pdiscard = blkdebug_co_pdiscard, - .bdrv_co_get_block_status = bdrv_co_get_block_status_from_file, + .bdrv_co_get_block_status = blkdebug_co_get_block_status, .bdrv_debug_event = blkdebug_debug_event, .bdrv_debug_breakpoint = blkdebug_debug_breakpoint,
Signature · 1 line
2.13.6
PATCH v6 23/24 qcow2: Relax is_zero() assertion
In reply to Eric BlakeNow that bdrv_is_allocated accepts non-aligned inputs, we can remove the TODO added in earlier refactoring.
---
v6: new patch [Kevin]
--- block/qcow2.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-)
block/qcow2.c +3 -9
@@ -3003,22 +3003,16 @@ static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) {int64_t nr;int res; -int64_t start; - -/* TODO: Widening to sector boundaries should only be needed as -* long as we can't query finer granularity. */ - start =QEMU_ALIGN_DOWN (offset,BDRV_SECTOR_SIZE ); - bytes =QEMU_ALIGN_UP (offset + bytes,BDRV_SECTOR_SIZE ) - start;/* Clamp to image length, before checking status of underlying sectors */ -if (start + bytes > bs->total_sectors *BDRV_SECTOR_SIZE ) { - bytes = bs->total_sectors *BDRV_SECTOR_SIZE - start; +if (offset + bytes > bs->total_sectors *BDRV_SECTOR_SIZE ) { + bytes = bs->total_sectors *BDRV_SECTOR_SIZE - offset; }if (!bytes) {return true; } - res =bdrv_block_status_above (bs,NULL , start, bytes, &nr,NULL ,NULL ); + res =bdrv_block_status_above (bs,NULL , offset, bytes, &nr,NULL ,NULL );return res >=0 && (res &BDRV_BLOCK_ZERO ) && nr == bytes; }
Signature · 1 line
2.13.6
PATCH v6 24/24 qemu-io: Relax 'alloc' now that block-status doesn't assert
In reply to Eric BlakePreviously, the alloc command required that input parameters be sector-aligned and clamped to 32 bits, because the underlying bdrv_is_allocated used a 32-bit parameter and asserted aligned inputs. But now that we have fixed block status to report a 64-bit bytes value, and to properly round requests on behalf of guests, we can pass any values, and can use qemu-io to add coverage that our rounding is correct regardless of the guest alignment constraints.
Update iotest 177 to intentionally probe block status at unaligned boundaries as well as with a bytes value that does not map to 32-bit sectors, which also required tweaking the image prep to leave an unallocated portion to the image under test.
---
v3: also test huge bytes value, R-b dropped
v2: new patch
--- qemu-io-cmds.c | 13 ------------- tests/qemu-iotests/177 | 12 ++++++++++-- tests/qemu-iotests/177.out | 19 ++++++++++++++----- 3 files changed, 24 insertions(+), 20 deletions(-)
qemu-io-cmds.c +0 -13
@@ -1769,10 +1769,6 @@ static int alloc_f(BlockBackend *blk, int argc, char **argv)if (offset <0 ) {print_cvtnum_err (offset, argv[1 ]);return 0 ; - }else if (!QEMU_IS_ALIGNED (offset,BDRV_SECTOR_SIZE )) { -printf ("%" PRId64" is not a sector-aligned value for 'offset'\n" , - offset); -return 0 ; }if (argc ==3 ) {
@@ -1780,19 +1776,10 @@ static int alloc_f(BlockBackend *blk, int argc, char **argv)if (count <0 ) {print_cvtnum_err (count, argv[2 ]);return 0 ; - }else if (count >INT_MAX *BDRV_SECTOR_SIZE ) { -printf ("length argument cannot exceed %llu, given %s\n" , -INT_MAX *BDRV_SECTOR_SIZE , argv[2 ]); -return 0 ; } }else { count =BDRV_SECTOR_SIZE ; } -if (!QEMU_IS_ALIGNED (count,BDRV_SECTOR_SIZE )) { -printf ("%" PRId64" is not a sector-aligned value for 'count'\n" , - count); -return 0 ; - } remaining = count; sum_alloc =0 ;
tests/qemu-iotests/177 +10 -2
@@ -51,7 +51,7 @@ echo "== setting up files ==" TEST_IMG="$TEST_IMG.base" _make_test_img $size $QEMU_IO -c "write -P 11 0 $size" "$TEST_IMG.base" | _filter_qemu_io _make_test_img -b "$TEST_IMG.base" -$QEMU_IO -c "write -P 22 0 $size" "$TEST_IMG" | _filter_qemu_io +$QEMU_IO -c "write -P 22 0 110M" "$TEST_IMG" | _filter_qemu_io # Limited to 64k max-transfer echo
@@ -82,6 +82,13 @@ $QEMU_IO -c "open -o $options,$limits blkdebug::$TEST_IMG" \ -c "discard 80000001 30M" | _filter_qemu_io echo +echo "== block status smaller than alignment ==" +limits=align=4k +$QEMU_IO -c "open -o $options,$limits blkdebug::$TEST_IMG" \ + -c "alloc 1 1" -c "alloc 0x6dffff0 1000" -c "alloc 127m 5P" \ + -c map | _filter_qemu_io + +echo echo "== verify image content ==" function verify_io()
@@ -103,7 +110,8 @@ function verify_io() echo read -P 0 32M 32M echo read -P 22 64M 13M echo read -P $discarded 77M 29M - echo read -P 22 106M 22M + echo read -P 22 106M 4M + echo read -P 11 110M 18M } verify_io | $QEMU_IO -r "$TEST_IMG" | _filter_qemu_io
tests/qemu-iotests/177.out +14 -5
@@ -5,8 +5,8 @@ Formatting 'TEST_DIR/t.IMGFMT.base', fmt=IMGFMT size=134217728 wrote 134217728/134217728 bytes at offset 0 128 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728 backing_file=TEST_DIR/t.IMGFMT.base -wrote 134217728/134217728 bytes at offset 0 -128 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +wrote 115343360/115343360 bytes at offset 0 +110 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) == constrained alignment and max-transfer == wrote 131072/131072 bytes at offset 1000
@@ -26,6 +26,13 @@ wrote 33554432/33554432 bytes at offset 33554432 discard 31457280/31457280 bytes at offset 80000001 30 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +== block status smaller than alignment == +1/1 bytes allocated at offset 1 bytes +16/1000 bytes allocated at offset 110 MiB +0/1048576 bytes allocated at offset 127 MiB +110 MiB (0x6e00000) bytes allocated at offset 0 bytes (0x0) +18 MiB (0x1200000) bytes not allocated at offset 110 MiB (0x6e00000) + == verify image content == read 1000/1000 bytes at offset 0 1000 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
@@ -43,12 +50,14 @@ read 13631488/13631488 bytes at offset 67108864 13 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) read 30408704/30408704 bytes at offset 80740352 29 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) -read 23068672/23068672 bytes at offset 111149056 -22 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +read 4194304/4194304 bytes at offset 111149056 +4 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +read 18874368/18874368 bytes at offset 115343360 +18 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) Offset Length File 0 0x800000 TEST_DIR/t.IMGFMT 0x900000 0x2400000 TEST_DIR/t.IMGFMT 0x3c00000 0x1100000 TEST_DIR/t.IMGFMT -0x6a00000 0x1600000 TEST_DIR/t.IMGFMT +0x6a00000 0x400000 TEST_DIR/t.IMGFMT No errors were found on the image. *** done
Signature · 1 line
2.13.6
Re: [Qemu-devel] [PATCH v6 23/24] qcow2: Relax is_zero() assertion
In reply to Eric BlakeQuoted 4 lines · Eric Blake
Now that
bdrv_is_allocatedaccepts non-aligned inputs, we can remove the TODO added in earlier refactoring.Signed-off-by Eric Blake <[email protected]>
I was a bit too hasty in the copy-and-paste from 22/24 - there is no assertion in this patch, so the subject would be better as:
qcow2: Reduce is_zero() rounding
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
Re: [Qemu-devel] [PATCH v6 07/24] block: Convert bdrv_get_block_status() to bytes
In reply to Eric BlakeAm 12.10.2017 um 05:47 hat Eric Blake geschrieben:
Quoted 60 lines
We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
Changing the name of the function from
bdrv_get_block_status()tobdrv_block_status()ensures that the compiler enforces that all callers are updated. For now, theio.clayer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers.There was an inherent limitation in returning the offset via the return value: we only have room for
BDRV_BLOCK_OFFSET_MASKbits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses.For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at
bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularlybdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors.For ease of review,
bdrv_get_block_status_above()will be tackled separately.Signed-off-by Eric Blake <[email protected]>---
v6: redo interface to return byte-based offset by reference [Kevin]
v5: drop pointless 'if (pnum)' [John], add comment
v4: no change
v3: clamp bytes to 32-bits, rather than asserting
v2: rebase to earlier changes--- include/block/block.h | 17 +++++++++-------- block/io.c | 47 ++++++++++++++++++++++++++++++++++------------- block/qcow2-cluster.c | 2 +- qemu-img.c | 25 ++++++++++++++----------- 4 files changed, 58 insertions(+), 33 deletions(-)include/block/block.h +1
-1@@ -121,7 +121,7 @@ typedef struct HDGeometry {#define BDRV_REQUEST_MAX_BYTES (BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS)/* -* Allocation status flags for bdrv_get_block_status() and friends. +* Allocation status flags for bdrv_block_status() and friends. * * Public flags: * BDRV_BLOCK_DATA: allocation for data at offset is tied to this layer @@ -136,10 +136,11 @@ typedef struct HDGeometry {* that the block layer recompute the answer from the returned
include/block/block.h typedef struct HDGeometry { · line 142
Quoted 10 lines
include/block/block.h +4
-4* BDS; must be accompanied by just BDRV_BLOCK_OFFSET_VALID. * -* If BDRV_BLOCK_OFFSET_VALID is set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK) -* represent the offset in the returned BDS that is allocated for the -* corresponding raw data; however, whether that offset actually contains -* data also depends on BDRV_BLOCK_DATA and BDRV_BLOCK_ZERO, as follows: +* If BDRV_BLOCK_OFFSET_VALID is set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK) of +* the return value (old interface) or the entire map parameter (new +* interface) represent the offset in the returned BDS that is allocated for +* the corresponding raw data.
Are there any functions using the old interface left at the end of the series or do we want a final patch that removes the old interface from the description?
Kevin
Re: [Qemu-devel] [PATCH v6 12/24] block: Convert bdrv_get_block_status_above() to bytes
In reply to Eric BlakeAm 12.10.2017 um 05:47 hat Eric Blake geschrieben:
Quoted 35 lines
We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
Changing the name of the function from
bdrv_get_block_status_above()tobdrv_block_status_above()ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, theio.clayer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers.For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at
bdrv_block_status(), plus updates for the new split return interface. But some code, particularlybdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computess->granularity>>BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to useis_power_of_2while there.For ease of review,
bdrv_get_block_status()was tackled separately.Signed-off-by Eric Blake <[email protected]>int bdrv_block_status(BlockDriverState *bs,
int64_toffset,int64_tbytes,
int64_t*pnum,int64_t*map, BlockDriverState **file)
{block/io.c +0
-1-int64_t ret;
block/io.c static int bdrv_common_block_status_above(BlockDriverState *bs, · line 2083
Quoted 10 lines
block/io.c +0
-10-int n; - -assert (QEMU_IS_ALIGNED (offset | bytes,BDRV_SECTOR_SIZE )); -assert (pnum); -/* -* The contract allows us to return pnum smaller than bytes, even -* if the next query would see the same status; we truncate the -* request to avoid overflowing the driver's 32-bit interface. -*/ - bytes =MIN (bytes,BDRV_REQUEST_MAX_BYTES );
Is the limitation to BDRV_REQUEST_MAX_BYTES going away without being replaced by a new one in bdrv_co_block_status()? What protects us now from 32-bit truncation?
Kevin
Re: [Qemu-devel] [PATCH v6 23/24] qcow2: Relax is_zero() assertion
In reply to Eric BlakeAm 12.10.2017 um 15:20 hat Eric Blake geschrieben:
Quoted 10 lines · Eric Blake
Quoted 4 lines
Now that
bdrv_is_allocatedaccepts non-aligned inputs, we can remove the TODO added in earlier refactoring.Signed-off-by Eric Blake <[email protected]>I was a bit too hasty in the copy-and-paste from 22/24 - there is no assertion in this patch, so the subject would be better as:
qcow2: Reduce
is_zero()rounding
In fact, I don't see any assertion being modified in patch 22 either.
Kevin
Re: [Qemu-devel] [PATCH v6 07/24] block: Convert bdrv_get_block_status() to bytes
In reply to Kevin WolfQuoted 18 lines · Kevin Wolf
Am 12.10.2017 um 05:47 hat Eric Blake geschrieben:
Quoted 7 lines
We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
Quoted 9 lines
*
- * IfBDRV_BLOCK_OFFSET_VALIDis set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK)
- * represent the offset in the returned BDS that is allocated for the
- * corresponding raw data; however, whether that offset actually contains
- * data also depends onBDRV_BLOCK_DATAandBDRV_BLOCK_ZERO, as follows:
+ * IfBDRV_BLOCK_OFFSET_VALIDis set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK) of
+ * the return value (old interface) or the entire map parameter (new
+ * interface) represent the offset in the returned BDS that is allocated for
+ * the corresponding raw data.
Quoted 3 lines
Are there any functions using the old interface left at the end of the series or do we want a final patch that removes the old interface from the description?
Done here, in series 4, where we get rid of the old interface in the drivers:
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg02941.html
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
Re: [Qemu-devel] [PATCH v6 12/24] block: Convert bdrv_get_block_status_above() to bytes
In reply to Kevin WolfQuoted 23 lines · Kevin Wolf
Am 12.10.2017 um 05:47 hat Eric Blake geschrieben:
Quoted 11 lines
We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
int bdrv_block_status(BlockDriverState *bs,
int64_toffset,int64_tbytes,
int64_t*pnum,int64_t*map, BlockDriverState **file)
{block/io.c +0
-1-int64_t ret;block/io.c static int bdrv_common_block_status_above(BlockDriverState *bs, · line 2083
Quoted 10 lines
block/io.c +0
-10-int n; - -assert (QEMU_IS_ALIGNED (offset | bytes,BDRV_SECTOR_SIZE )); -assert (pnum); -/* -* The contract allows us to return pnum smaller than bytes, even -* if the next query would see the same status; we truncate the -* request to avoid overflowing the driver's 32-bit interface. -*/ - bytes =MIN (bytes,BDRV_REQUEST_MAX_BYTES );
Quoted 3 lines
Is the limitation to
BDRV_REQUEST_MAX_BYTESgoing away without being replaced by a new one inbdrv_co_block_status()? What protects us now from 32-bit truncation?
As of this patch, we have this 64-bit-clean call chain:bdrv_block_status()
bdrv_block_status_above()
bdrv_common_block_status_above()
bdrv_block_status_above_co_entry()
bdrv_co_block_status_above()
bdrv_co_block_status()
which in turn has:
/*
* The contract allows us to return pnum smaller than bytes, even
* if the next query would see the same status; we truncate the
* request to avoid overflowing the driver's 32-bit interface. */
bytes = MIN(bytes, BDRV_REQUEST_MAX_BYTES);
added earlier in 8/24.
At the end of series 3, the truncation prevention is still in place when calling into the drivers, per the changes to bdrv_co_block_status() in 21/24.
Later, in series 4, all drivers are converted to 64-bit interfaces, where the drivers are either made 64-bit clean, or do their own truncation to 32 bits.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
Re: [Qemu-devel] [PATCH v6 07/24] block: Convert bdrv_get_block_status() to bytes
In reply to Eric BlakeAm 20.10.2017 um 17:12 hat Eric Blake geschrieben:
Quoted 19 lines · Kevin Wolf
Quoted 18 lines
Am 12.10.2017 um 05:47 hat Eric Blake geschrieben:
Quoted 16 lines
We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
*
- * IfBDRV_BLOCK_OFFSET_VALIDis set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK)
- * represent the offset in the returned BDS that is allocated for the
- * corresponding raw data; however, whether that offset actually contains
- * data also depends onBDRV_BLOCK_DATAandBDRV_BLOCK_ZERO, as follows:
+ * IfBDRV_BLOCK_OFFSET_VALIDis set, bits 9-62 (BDRV_BLOCK_OFFSET_MASK) of
+ * the return value (old interface) or the entire map parameter (new
+ * interface) represent the offset in the returned BDS that is allocated for
+ * the corresponding raw data.
Quoted 7 lines
Quoted 3 lines
Are there any functions using the old interface left at the end of the series or do we want a final patch that removes the old interface from the description?
Done here, in series 4, where we get rid of the old interface in the drivers:
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg02941.html
Oh, right, I forgot about the driver callbacks. Makes sense.
Kevin
Re: [Qemu-devel] [PATCH v6 23/24] qcow2: Relax is_zero() assertion
In reply to Kevin WolfQuoted 5 lines · Kevin Wolf
Am 12.10.2017 um 15:20 hat Eric Blake geschrieben:
Quoted 4 lines · Eric Blake
Quoted 3 lines
Now that
bdrv_is_allocatedaccepts non-aligned inputs, we can remove the TODO added in earlier refactoring.
Quoted 8 lines
Quoted 6 lines
Quoted 1 line
Signed-off-by Eric Blake <[email protected]>I was a bit too hasty in the copy-and-paste from 22/24 - there is no assertion in this patch, so the subject would be better as:
qcow2: Reduce
is_zero()roundingIn fact, I don't see any assertion being modified in patch 22 either.
Oh, indeed. Not sure what I was thinking; I guess it's because I added 22 at the same time I added several other TODO assertions to series 1 a couple months ago, with the plans to relax rounding/assertions as appropriate later in the series, and then got confused on which things were being relaxed where ;)
If it's just the subject lines for 22 and 23 that need fixing, I trust you can do that as maintainer.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
Re: [Qemu-devel] [PATCH v6 12/24] block: Convert bdrv_get_block_status_above() to bytes
In reply to Eric BlakeAm 20.10.2017 um 17:22 hat Eric Blake geschrieben:
Quoted 44 lines · Kevin Wolf
Quoted 23 lines
Am 12.10.2017 um 05:47 hat Eric Blake geschrieben:
Quoted 11 lines
We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access.
int bdrv_block_status(BlockDriverState *bs,
int64_toffset,int64_tbytes,
int64_t*pnum,int64_t*map, BlockDriverState **file)
{block/io.c +0
-1-int64_t ret;block/io.c static int bdrv_common_block_status_above(BlockDriverState *bs, · line 2083
Quoted 10 lines
block/io.c +0
-10-int n; - -assert (QEMU_IS_ALIGNED (offset | bytes,BDRV_SECTOR_SIZE )); -assert (pnum); -/* -* The contract allows us to return pnum smaller than bytes, even -* if the next query would see the same status; we truncate the -* request to avoid overflowing the driver's 32-bit interface. -*/ - bytes =MIN (bytes,BDRV_REQUEST_MAX_BYTES );Quoted 3 lines
Is the limitation to
BDRV_REQUEST_MAX_BYTESgoing away without being replaced by a new one inbdrv_co_block_status()? What protects us now from 32-bit truncation?As of this patch, we have this 64-bit-clean call chain:
bdrv_block_status()
bdrv_block_status_above()
bdrv_common_block_status_above()
bdrv_block_status_above_co_entry()
bdrv_co_block_status_above()
bdrv_co_block_status()which in turn has:
/*
* The contract allows us to return pnum smaller than bytes, even
* if the next query would see the same status; we truncate the
* request to avoid overflowing the driver's 32-bit interface. */
bytes =MIN(bytes, BDRV_REQUEST_MAX_BYTES);
Quoted 5 lines
added earlier in 8/24.
At the end of series 3, the truncation prevention is still in place when calling into the drivers, per the changes to
bdrv_co_block_status()in 21/24.
I looked at the final state and missed it because now INT_MAX is used rather than BDRV_REQUEST_MAX_BYTES. Anyway, seems to be okay then.
Kevin
Kevin Wolf
AppliedAm 12.10.2017 um 05:46 hat Eric Blake geschrieben:
Quoted 7 lines
There are patches floating around to add
NBD_CMD_BLOCK_STATUS, but NBD wants to report status on byte granularity (even if the reporting will probably be naturally aligned to sectors or even much higher levels). I've therefore started the task of converting our block status code to report at a byte granularity rather than sectors.
Quoted 7 lines
Now that 2.11 is open, I'm rebasing/reposting the remaining patches.
The overall conversion currently looks like:
part 1:bdrv_is_allocated(merged, commit51b0a488)
part 2: dirty-bitmap (merged, commitca759622)
part 3:bdrv_get_block_status(this series, v5 at [1])
part 4: .bdrv_co_block_status (v3 is posted [2], mostly reviewed, but needs rebase)
Thanks, applied to the block branch.
Kevin
Quoted 16 lines · Kevin Wolf
Am 12.10.2017 um 05:46 hat Eric Blake geschrieben:
Quoted 7 lines
There are patches floating around to add
NBD_CMD_BLOCK_STATUS, but NBD wants to report status on byte granularity (even if the reporting will probably be naturally aligned to sectors or even much higher levels). I've therefore started the task of converting our block status code to report at a byte granularity rather than sectors.Quoted 7 lines
Now that 2.11 is open, I'm rebasing/reposting the remaining patches.
The overall conversion currently looks like:
part 1:bdrv_is_allocated(merged, commit51b0a488)
part 2: dirty-bitmap (merged, commitca759622)
part 3:bdrv_get_block_status(this series, v5 at [1])
part 4: .bdrv_co_block_status (v3 is posted [2], mostly reviewed, but needs rebase)
Quoted 4 lines
Thanks, applied to the block branch.
Kevin
Thanks;
Eric: does this just leave part four?
Quoted 6 lines · John Snow
Quoted 6 lines
Quoted 5 lines
The overall conversion currently looks like:
part 1:bdrv_is_allocated(merged, commit51b0a488)
part 2: dirty-bitmap (merged, commitca759622)
part 3:bdrv_get_block_status(this series, v5 at [1])
part 4: .bdrv_co_block_status (v3 is posted [2], mostly reviewed, but needs rebase)
Quoted 8 lines
Quoted 4 lines
Thanks, applied to the block branch.
Kevin
Thanks;
Eric: does this just leave part four?
Yes; the latest version of part four is v4:
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg02940.html
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org