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

flexnbd: Remove the server I/O lock from around NBD requests

Browse Source 
NBD doesn't actually guarantee what happens if you have two
concurrent writes to overlapping areas of the disc, and this
mutex was causing us a near-deadlock when the TCP connection
died uncleanly, partway through a request. So now we don't
bother. This actually removes the last user of the server I/O
mutex, so we can remove it completely from the codebase in a
future commit.
This commit is contained in:
nick
2013-09-23 10:22:48 +01:00
parent 71036730c4
commit 9a3106f946
1 changed files with 0 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/client.c
View File

@@ -560,7 +560,6 @@ int client_serve_request(struct client* client)
return client->disconnect;
}
server_lock_io( client->serve );
{
if ( !server_is_closed( client->serve ) ) {
/* We arm / disarm around client_reply() to catch cases where the
@@ -581,7 +580,6 @@ int client_serve_request(struct client* client)
stop = 0;
}
}
server_unlock_io( client->serve );
return stop;
@@ -616,7 +614,6 @@ void client_cleanup(struct client* client,
client->fileno = -1;
}
if ( server_io_locked( client->serve ) ) { server_unlock_io( client->serve ); }
if ( server_acl_locked( client->serve ) ) { server_unlock_acl( client->serve ); }
}