bytemark/flexnbd-c
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Block allocation map is now built in a separate thread, and does not delay

Browse Source 
server startup (sparse write avoidance doesn't happen until it is finished).
Added mutex to bitset functions, which were already being called from
multiple threads.  Rewrote allocation map builder to request file
information in multiple chunks, to avoid uninterruptible wait and dynamic
memory allocation.
This commit is contained in:
Matthew Bloch
2012-10-07 21:55:01 +01:00
parent 7b13964c39
commit a49cf14927
6 changed files with 112 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/client.c
View File

@@ -407,13 +407,11 @@ void client_reply_to_read( struct client* client, struct nbd_request request )
void client_reply_to_write( struct client* client, struct nbd_request request )
{
debug("request write %ld+%d", request.from, request.len);
if (client->serve->allocation_map) {
if (client->serve->allocation_map_built) {
write_not_zeroes( client, request.from, request.len );
}
else {
debug("No allocation map, writing directly.");
/* If we get cut off partway through reading this data
* */
ERROR_IF_NEGATIVE(
readloop( client->socket,
client->mapped + request.from,
@@ -423,6 +421,14 @@ void client_reply_to_write( struct client* client, struct nbd_request request )
request.len
);
server_dirty(client->serve, request.from, request.len);
/* the allocation_map is shared between client threads, and may be
* being built. But AFAICT this is safe, to accurately reflect the
* fact that we've caused block allocation to occur, though we will
* never consult the allocation_map until the builder thread has
* finished.
*/
bitset_set_range(client->serve->allocation_map,
request.from, request.len);
}
if (1) /* not sure whether this is necessary... */