Hi all!
It's an alternative lock-less solution to
[PATCH v4 0/3] qcow2: fix parallel rewrite and discard (rw-lock)
It's an updated version of what was don in
[PATCH v3 0/6] qcow2: compressed write cache
I've split the logic into several patches, add a lot more comments and
documentation, I hope it's all now a lot more clean.
Main change from v3 is careful handling "freeing" the cluster. The
concept of "free cluster" is established. And good thing: now updating
inflight-counters it lock-less: we don't need the lock for discard
operation, instead we consider this discard as the last one inflight
write!
The solution is still more complicated then co-rw-lock based one. But it
is cool and I hope someone like it.
RFC: should we protect reads as well of intersecting with discarding
host clusters? I think yes. Reads are not as critical, as they can't
corrupt the metadata or data (due to use-after-free like writes), but
with reads it's possible that guest reads some metadata cluster. It's a
kind of security violation I think.
Vladimir Sementsov-Ogievskiy (6):
iotests: add qcow2-discard-during-rewrite
qcow2: fix cache discarding in update_refcount()
qcow2: introduce is_cluster_free() helper
qcow2: introduce inflight-write-counters
qcow2: consider in-flight-writes when freeing clusters
qcow2: do not discard host clusters during in-flight writes
block/qcow2.h | 16 ++
block/qcow2-refcount.c | 184 +++++++++++++++++-
block/qcow2.c | 26 ++-
.../tests/qcow2-discard-during-rewrite | 72 +++++++
.../tests/qcow2-discard-during-rewrite.out | 21 ++
5 files changed, 308 insertions(+), 11 deletions(-)
create mode 100755 tests/qemu-iotests/tests/qcow2-discard-during-rewrite
create mode 100644 tests/qemu-iotests/tests/qcow2-discard-during-rewrite.out
--
2.29.2