Message ID | 1423671788-31716-1-git-send-email-kwolf@redhat.com |
---|---|
State | New |
Headers | show |
Am 11.02.2015 um 17:23 hat Kevin Wolf geschrieben: > This implements bdrv_co_get_status() for VHD images. This can s/bdrv_co_get_status/bdrv_co_get_block_status/, obviously (in the subject as well). > significantly speed up qemu-img convert operation because only with this > function implemented sparseness can be considered. (Before, converting a > 1 TB empty image took several minutes for me, now it's instantaneous.) > > Signed-off-by: Kevin Wolf <kwolf@redhat.com> Kevin
On 2015-02-11 at 11:23, Kevin Wolf wrote: > This implements bdrv_co_get_status() for VHD images. This can > significantly speed up qemu-img convert operation because only with this > function implemented sparseness can be considered. (Before, converting a > 1 TB empty image took several minutes for me, now it's instantaneous.) > > Signed-off-by: Kevin Wolf <kwolf@redhat.com> > --- > block/vpc.c | 54 ++++++++++++++++++++++++++++++++++++++++++++++++++++-- > 1 file changed, 52 insertions(+), 2 deletions(-) > > diff --git a/block/vpc.c b/block/vpc.c > index 46803b1..379ff4a 100644 > --- a/block/vpc.c > +++ b/block/vpc.c > @@ -597,6 +597,55 @@ static coroutine_fn int vpc_co_write(BlockDriverState *bs, int64_t sector_num, > return ret; > } > > +static int64_t coroutine_fn vpc_co_get_block_status(BlockDriverState *bs, > + int64_t sector_num, int nb_sectors, int *pnum) > +{ > + BDRVVPCState *s = bs->opaque; > + VHDFooter *footer = (VHDFooter *) s->footer_buf; Double space before the left parenthesis. > + int64_t start, offset; > + bool allocated; > + int n; > + > + if (be32_to_cpu(footer->type) == VHD_FIXED) { > + *pnum = nb_sectors; > + return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA | > + (sector_num << BDRV_SECTOR_BITS); > + } > + > + offset = get_sector_offset(bs, sector_num, 0); > + > + *pnum = 1; > + start = offset; > + allocated = (offset != -1); > + nb_sectors--; > + > + while (nb_sectors > 0) { > + offset = get_sector_offset(bs, sector_num, 0); During the first iteration, this will is exactly the function call as the one before this loop. > + > + if ((offset != -1) != allocated) { > + break; > + } > + > + if (allocated && offset != start + (*pnum * 512)) { Therefore, if allocated is true, this check will always fail because offset == start. > + break; > + } > + > + /* All sectors in a block are contiguous (without using the bitmap) */ > + n = ROUND_UP(sector_num + 1, s->block_size / 512) - sector_num; > + n = MIN(n, nb_sectors); I think you can make use of this assumption even before calling get_sector_offset() in the first iteration. Also, 512 is shorter than BDRV_SECTOR_SIZE and it's nice to get it on a single line, but I think not introducing another literal 512 in the code would be nicer; the same applies above in "*pnum * 512". Max > + > + *pnum += n; > + sector_num += n; > + nb_sectors -= n; > + } > + > + if (allocated) { > + return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start; > + } else { > + return 0; > + } > +} > + > /* > * Calculates the number of cylinders, heads and sectors per cylinder > * based on a given number of sectors. This is the algorithm described > @@ -907,8 +956,9 @@ static BlockDriver bdrv_vpc = { > .bdrv_reopen_prepare = vpc_reopen_prepare, > .bdrv_create = vpc_create, > > - .bdrv_read = vpc_co_read, > - .bdrv_write = vpc_co_write, > + .bdrv_read = vpc_co_read, > + .bdrv_write = vpc_co_write, > + .bdrv_co_get_block_status = vpc_co_get_block_status, > > .bdrv_get_info = vpc_get_info, >
Am 11.02.2015 um 23:51 hat Max Reitz geschrieben: > On 2015-02-11 at 11:23, Kevin Wolf wrote: > >This implements bdrv_co_get_status() for VHD images. This can > >significantly speed up qemu-img convert operation because only with this > >function implemented sparseness can be considered. (Before, converting a > >1 TB empty image took several minutes for me, now it's instantaneous.) > > > >Signed-off-by: Kevin Wolf <kwolf@redhat.com> > >--- > > block/vpc.c | 54 ++++++++++++++++++++++++++++++++++++++++++++++++++++-- > > 1 file changed, 52 insertions(+), 2 deletions(-) > > > >diff --git a/block/vpc.c b/block/vpc.c > >index 46803b1..379ff4a 100644 > >--- a/block/vpc.c > >+++ b/block/vpc.c > >@@ -597,6 +597,55 @@ static coroutine_fn int vpc_co_write(BlockDriverState *bs, int64_t sector_num, > > return ret; > > } > >+static int64_t coroutine_fn vpc_co_get_block_status(BlockDriverState *bs, > >+ int64_t sector_num, int nb_sectors, int *pnum) > >+{ > >+ BDRVVPCState *s = bs->opaque; > >+ VHDFooter *footer = (VHDFooter *) s->footer_buf; > > Double space before the left parenthesis. Fun. As it happens, I copied this line from elsewhere... > >+ int64_t start, offset; > >+ bool allocated; > >+ int n; > >+ > >+ if (be32_to_cpu(footer->type) == VHD_FIXED) { > >+ *pnum = nb_sectors; > >+ return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA | > >+ (sector_num << BDRV_SECTOR_BITS); > >+ } > >+ > >+ offset = get_sector_offset(bs, sector_num, 0); > >+ > >+ *pnum = 1; > >+ start = offset; > >+ allocated = (offset != -1); > >+ nb_sectors--; > >+ > >+ while (nb_sectors > 0) { > >+ offset = get_sector_offset(bs, sector_num, 0); > > During the first iteration, this will is exactly the function call > as the one before this loop. > > >+ > >+ if ((offset != -1) != allocated) { > >+ break; > >+ } > >+ > >+ if (allocated && offset != start + (*pnum * 512)) { > > Therefore, if allocated is true, this check will always fail because > offset == start. Whoops, thanks for catching that. I'll fix it, even though it doesn't seem to have any practicel consequences. VHD is a format braindead enough that you never get two contiguous blocks anyway. There's always a bitmap in between. > >+ break; > >+ } > >+ > >+ /* All sectors in a block are contiguous (without using the bitmap) */ > >+ n = ROUND_UP(sector_num + 1, s->block_size / 512) - sector_num; > >+ n = MIN(n, nb_sectors); > > I think you can make use of this assumption even before calling > get_sector_offset() in the first iteration. Hm... Maybe I need to make it a do-while loop to get rid of some duplication with the first iteration. > Also, 512 is shorter than BDRV_SECTOR_SIZE and it's nice to get it > on a single line, but I think not introducing another literal 512 in > the code would be nicer; the same applies above in "*pnum * 512". Fair enough. Kevin
diff --git a/block/vpc.c b/block/vpc.c index 46803b1..379ff4a 100644 --- a/block/vpc.c +++ b/block/vpc.c @@ -597,6 +597,55 @@ static coroutine_fn int vpc_co_write(BlockDriverState *bs, int64_t sector_num, return ret; } +static int64_t coroutine_fn vpc_co_get_block_status(BlockDriverState *bs, + int64_t sector_num, int nb_sectors, int *pnum) +{ + BDRVVPCState *s = bs->opaque; + VHDFooter *footer = (VHDFooter *) s->footer_buf; + int64_t start, offset; + bool allocated; + int n; + + if (be32_to_cpu(footer->type) == VHD_FIXED) { + *pnum = nb_sectors; + return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA | + (sector_num << BDRV_SECTOR_BITS); + } + + offset = get_sector_offset(bs, sector_num, 0); + + *pnum = 1; + start = offset; + allocated = (offset != -1); + nb_sectors--; + + while (nb_sectors > 0) { + offset = get_sector_offset(bs, sector_num, 0); + + if ((offset != -1) != allocated) { + break; + } + + if (allocated && offset != start + (*pnum * 512)) { + break; + } + + /* All sectors in a block are contiguous (without using the bitmap) */ + n = ROUND_UP(sector_num + 1, s->block_size / 512) - sector_num; + n = MIN(n, nb_sectors); + + *pnum += n; + sector_num += n; + nb_sectors -= n; + } + + if (allocated) { + return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start; + } else { + return 0; + } +} + /* * Calculates the number of cylinders, heads and sectors per cylinder * based on a given number of sectors. This is the algorithm described @@ -907,8 +956,9 @@ static BlockDriver bdrv_vpc = { .bdrv_reopen_prepare = vpc_reopen_prepare, .bdrv_create = vpc_create, - .bdrv_read = vpc_co_read, - .bdrv_write = vpc_co_write, + .bdrv_read = vpc_co_read, + .bdrv_write = vpc_co_write, + .bdrv_co_get_block_status = vpc_co_get_block_status, .bdrv_get_info = vpc_get_info,
This implements bdrv_co_get_status() for VHD images. This can significantly speed up qemu-img convert operation because only with this function implemented sparseness can be considered. (Before, converting a 1 TB empty image took several minutes for me, now it's instantaneous.) Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- block/vpc.c | 54 ++++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 52 insertions(+), 2 deletions(-)