On 3/5/21 5:31 PM, Eric Blake wrote:
This is another API/ABI change to filters, but that doesn't matter. Instead of having a 1:1 correspondence of automatic backend context to the current connection (which means filters.c has to consult GET_CONN on every action), switch things so that a given context tracks the next context. This requires another signature change to the internal backend_open and backend_reopen functions, as well as a public change to the type of the argument handed to filter .open calls. A given connection now only needs to track the top_context, and we no longer need get_context()/set_context(). Internally, nbdkit_context and nbdkit_next are the same type, but the public interface for filters keeps them separate for type-safety reasons (assigning a backend handle to the current filter's context is different than calling into the backend via next_ops/nxdata as the backend's context).
@@ -471,7 +471,7 @@ outer filter to the plugin will be in reverse. Skipping a call to C<next> is acceptable if the filter will not access C<next_ops> during any of the remaining callbacks reached on the same connection.
-The value of C<backend> in this call has a lifetime that lasts until +The value of C<context> in this call has a lifetime that lasts until the counterpart C<.close>, and it is this value (and not the distinct C<nxdata> of C<.pread> and friends) that must be passed as the first parameter to C<nbdkit_backend_reopen> by a filter attempting to retry @@ -481,14 +481,14 @@ a connection into the backend.
Filters have access to a function for reopening the backend:
- int (nbdkit_backend *backend, int readonly, const char *exportname, + int (nbdkit_context *context, int readonly, const char *exportname, void **nxdata);
This goes away in a later patch, but I might as well spell it correctly in the earlier patches: s/int /int nbdkit_backend_reopen)/ -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org