Re: [Qemu-devel] [PATCH v3 11/19] block: define get_block_status return value

[Kevin Wolf]

Am 30.07.2013 um 16:19 hat Paolo Bonzini geschrieben:
> Il 30/07/2013 16:14, Kevin Wolf ha scritto:
> > Am 25.07.2013 um 16:23 hat Paolo Bonzini geschrieben:
> >> Define the return value of get_block_status.  Bits 0, 1, 2 and 9-62
> >> are valid; bit 63 (the sign bit) is reserved for errors.  Bits 3-8
> >> are left for future extensions.
> >>
> >> The return code is compatible with the old is_allocated API: if a driver
> >> only returns 0 or 1 (aka BDRV_BLOCK_DATA) like is_allocated used to,
> >> clients of is_allocated will not have any change in behavior.  Still,
> >> we will return more precise information in the next patches and the
> >> new definition of bdrv_is_allocated is already prepared for this.
> >>
> >> Reviewed-by: Eric Blake <address@hidden>
> >> Signed-off-by: Paolo Bonzini <address@hidden>
> >> ---
> >>  block.c               | 10 ++++++++--
> >>  include/block/block.h | 26 ++++++++++++++++++++++++++
> >>  2 files changed, 34 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/block.c b/block.c
> >> index f533c36..7cfbf71 100644
> >> --- a/block.c
> >> +++ b/block.c
> >> @@ -3004,7 +3004,7 @@ static int64_t coroutine_fn 
> >> bdrv_co_get_block_status(BlockDriverState *bs,
> >>  
> >>      if (!bs->drv->bdrv_co_get_block_status) {
> >>          *pnum = nb_sectors;
> >> -        return 1;
> >> +        return BDRV_BLOCK_DATA;
> >>      }
> >>  
> >>      return bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, 
> >> pnum);
> >> @@ -3054,7 +3054,13 @@ int64_t bdrv_get_block_status(BlockDriverState *bs, 
> >> int64_t sector_num,
> >>  int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t 
> >> sector_num,
> >>                                     int nb_sectors, int *pnum)
> >>  {
> >> -    return bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
> >> +    int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
> >> +    if (ret < 0) {
> >> +        return ret;
> >> +    }
> >> +    return
> >> +        (ret & BDRV_BLOCK_DATA) ||
> >> +        ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs));
> >>  }
> >>  
> >>  /*
> >> diff --git a/include/block/block.h b/include/block/block.h
> >> index e41854e..d044b31 100644
> >> --- a/include/block/block.h
> >> +++ b/include/block/block.h
> >> @@ -81,6 +81,32 @@ typedef struct BlockDevOps {
> >>  #define BDRV_SECTOR_SIZE   (1ULL << BDRV_SECTOR_BITS)
> >>  #define BDRV_SECTOR_MASK   ~(BDRV_SECTOR_SIZE - 1)
> >>  
> >> +/* BDRV_BLOCK_DATA: data is read from bs->file or another file
> >> + * BDRV_BLOCK_ZERO: sectors read as zero
> >> + * BDRV_BLOCK_OFFSET_VALID: sector stored in bs->file as raw data
> >> + *
> >> + * If BDRV_BLOCK_OFFSET_VALID is set, bits 9-62 represent the offset in
> >> + * bs->file where sector data can be read from as raw data.
> >> + *
> >> + * DATA == 0 && ZERO == 0 means that data is read from backing_hd if 
> >> present.
> >> + *
> >> + * DATA ZERO OFFSET_VALID
> >> + *  t    t        t       sectors read as zero, bs->file is zero at offset
> >> + *  t    f        t       sectors read as valid from bs->file at offset
> >> + *  f    t        t       sectors preallocated, read as zero, bs->file not
> >> + *                        necessarily zero at offset
> >> + *  f    f        t       sectors preallocated but read from backing_hd,
> >> + *                        bs->file contains garbage at offset
> >> + *  t    t        f       sectors preallocated, read as zero, unknown 
> >> offset
> >> + *  t    f        f       sectors read from unknown file or offset
> >> + *  f    t        f       not allocated or unknown offset, read as zero
> >> + *  f    f        f       not allocated or unknown offset, read from 
> >> backing_hd
> >> + */
> >> +#define BDRV_BLOCK_DATA         1
> >> +#define BDRV_BLOCK_ZERO         2
> >> +#define BDRV_BLOCK_OFFSET_VALID 4
> >> +#define BDRV_BLOCK_OFFSET_MASK  BDRV_SECTOR_MASK
> > 
> > When are block driver supposed to set the BDRV_BLOCK_OFFSET_VALID flag?
> 
> Always if they can provide the information.
> 
> > For example, qcow2 could in theory set the flag, it has all of the
> > information already in memory.
> 
> In fact it does later in the series.
> 
> > But with a fragmented image this might
> > mean that it returns only one cluster instead of a large area with one
> > bdrv_get_block_status() call.
> 
> This is just theoretical, right?  qcow2_get_cluster_offset only works on
> areas that are contiguous in the raw image.

Right, somehow I was thinking that .bdrv_is_allocated() used something
more clever, but it doesn't. So this isn't worse than what we have today.

Kevin