We have max_transfer documented in BlockLimits, but while we honor it during pwrite_zeroes, we were blindly ignoring it during pwritev and preadv, leading to multiple drivers having to implement fragmentation themselves. This series moves fragmentation to the block layer, then fixes the NBD and iscsi driver to use it.
qcow2 still does self-fragmenting, but that's because of cluster boundaries where it really has to do additional work beyond what the block layer can automatically provide. Prequisite: Kevin's latest block branch PULL request Also available as a tag at: git fetch git://repo.or.cz/qemu/ericb.git nbd-fragment-v3 Changes since v2: - patch 1, 3: change return semantics to be 0, not bytes, on success, since at least one caller asserts during 'make check' otherwise 001/6:[0008] [FC] 'block: Fragment reads to max transfer length' 002/6:[----] [--] 'raw_bsd: Don't advertise flags not supported by protocol layer' 003/6:[0002] [FC] 'block: Fragment writes to max transfer length' 004/6:[----] [--] 'nbd: Rely on block layer to break up large requests' 005/6:[----] [--] 'nbd: Drop unused offset parameter' 006/6:[----] [--] 'iscsi: Rely on block layer to break up large requests' Eric Blake (6): block: Fragment reads to max transfer length raw_bsd: Don't advertise flags not supported by protocol layer block: Fragment writes to max transfer length nbd: Rely on block layer to break up large requests nbd: Drop unused offset parameter iscsi: Rely on block layer to break up large requests include/block/nbd.h | 1 - nbd/nbd-internal.h | 4 +-- block/io.c | 90 +++++++++++++++++++++++++++++++++++++++-------------- block/iscsi.c | 14 +++------ block/nbd-client.c | 78 +++++++++++++--------------------------------- block/nbd.c | 12 ++----- block/raw_bsd.c | 6 ++-- nbd/common.c | 5 +-- 8 files changed, 103 insertions(+), 107 deletions(-) -- 2.5.5
