block: Generalise and optimise COR serialisation
Change the API so that specific requests can be marked serialising. Only
these requests are checked for overlaps then.
This means that during a Copy on Read operation, not all requests
overlapping other requests are serialised any more, but only those that
actually overlap with the specific COR request.
Also remove COR from function and variable names because this
functionality can be useful in other contexts.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
diff --git a/block.c b/block.c
index 0a391bd..493c75f 100644
--- a/block.c
+++ b/block.c
@@ -2208,6 +2208,10 @@
*/
static void tracked_request_end(BdrvTrackedRequest *req)
{
+ if (req->serialising) {
+ req->bs->serialising_in_flight--;
+ }
+
QLIST_REMOVE(req, list);
qemu_co_queue_restart_all(&req->wait_queue);
}
@@ -2222,10 +2226,11 @@
{
*req = (BdrvTrackedRequest){
.bs = bs,
- .offset = offset,
- .bytes = bytes,
- .is_write = is_write,
- .co = qemu_coroutine_self(),
+ .offset = offset,
+ .bytes = bytes,
+ .is_write = is_write,
+ .co = qemu_coroutine_self(),
+ .serialising = false,
};
qemu_co_queue_init(&req->wait_queue);
@@ -2233,6 +2238,14 @@
QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
}
+static void mark_request_serialising(BdrvTrackedRequest *req)
+{
+ if (!req->serialising) {
+ req->bs->serialising_in_flight++;
+ req->serialising = true;
+ }
+}
+
/**
* Round a region to cluster boundaries
*/
@@ -2285,26 +2298,31 @@
return true;
}
-static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
- BdrvTrackedRequest *self, int64_t offset, unsigned int bytes)
+static void coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
{
+ BlockDriverState *bs = self->bs;
BdrvTrackedRequest *req;
int64_t cluster_offset;
unsigned int cluster_bytes;
bool retry;
+ if (!bs->serialising_in_flight) {
+ return;
+ }
+
/* If we touch the same cluster it counts as an overlap. This guarantees
* that allocating writes will be serialized and not race with each other
* for the same cluster. For example, in copy-on-read it ensures that the
* CoR read and write operations are atomic and guest writes cannot
* interleave between them.
*/
- round_bytes_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
+ round_bytes_to_clusters(bs, self->offset, self->bytes,
+ &cluster_offset, &cluster_bytes);
do {
retry = false;
QLIST_FOREACH(req, &bs->tracked_requests, list) {
- if (req == self) {
+ if (req == self || (!req->serialising && !self->serialising)) {
continue;
}
if (tracked_request_overlaps(req, cluster_offset, cluster_bytes)) {
@@ -2923,12 +2941,10 @@
/* Handle Copy on Read and associated serialisation */
if (flags & BDRV_REQ_COPY_ON_READ) {
- bs->copy_on_read_in_flight++;
+ mark_request_serialising(req);
}
- if (bs->copy_on_read_in_flight) {
- wait_for_overlapping_requests(bs, req, offset, bytes);
- }
+ wait_serialising_requests(req);
if (flags & BDRV_REQ_COPY_ON_READ) {
int pnum;
@@ -2977,10 +2993,6 @@
}
out:
- if (flags & BDRV_REQ_COPY_ON_READ) {
- bs->copy_on_read_in_flight--;
- }
-
return ret;
}
@@ -3179,9 +3191,7 @@
assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
- if (bs->copy_on_read_in_flight) {
- wait_for_overlapping_requests(bs, req, offset, bytes);
- }
+ wait_serialising_requests(req);
ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
