search for: plugin_lock_request

...in the crashing case. AFAICS the python API says that once you have threads, you have to do the GIL dance: https://docs.python.org/2/c-api/init.html#thread-state-and-the-global-interpreter-lock . All the crashes I saw can be explained by unsynchronized accesses to python objects. I found the bug. plugin_lock_request() is used to lock client requests only, but it does _not_ lock any other calls into the plugin. Due to that, the following calls are neither locked against each other nor are they locked against client requests: plugin_register (const char *_filename, plugin_cleanup (void) plugin_config (const char...

...ckin, int sockout) { int r; - plugin_lock_connection (); + lock_connection (); r = _handle_single_connection (sockin, sockout); - plugin_unlock_connection (); + unlock_connection (); return r; } @@ -740,12 +740,12 @@ negotiate_handshake (struct connection *conn) { int r; - plugin_lock_request (conn); + lock_request (conn); if (!newstyle) r = _negotiate_handshake_oldstyle (conn); else r = _negotiate_handshake_newstyle (conn); - plugin_unlock_request (conn); + unlock_request (conn); return r; } @@ -1057,9 +1057,9 @@ recv_request_send_reply (struct connection *conn...

...ckin, int sockout) { int r; - plugin_lock_connection (); + lock_connection (); r = _handle_single_connection (sockin, sockout); - plugin_unlock_connection (); + unlock_connection (); return r; } @@ -740,12 +740,12 @@ negotiate_handshake (struct connection *conn) { int r; - plugin_lock_request (conn); + lock_request (conn); if (!newstyle) r = _negotiate_handshake_oldstyle (conn); else r = _negotiate_handshake_newstyle (conn); - plugin_unlock_request (conn); + unlock_request (conn); return r; } @@ -1057,9 +1057,9 @@ recv_request_send_reply (struct connection *conn...

...n_unlock_connection (void) { - assert (dl); - if (plugin._thread_model <= NBDKIT_THREAD_MODEL_SERIALIZE_CONNECTIONS) { debug ("%s: release connection lock", filename); pthread_mutex_unlock (&connection_lock); @@ -321,8 +325,6 @@ plugin_unlock_connection (void) void plugin_lock_request (struct connection *conn) { - assert (dl); - if (plugin._thread_model <= NBDKIT_THREAD_MODEL_SERIALIZE_ALL_REQUESTS) { debug ("acquire global request lock"); pthread_mutex_lock (&all_requests_lock); @@ -332,12 +334,16 @@ plugin_lock_request (struct connection *conn)...

I want to make my nbd forwarding plugin fully parallel - but to do that, I first need to make nbdkit itself fully parallel ;) With this series, I was finally able to demonstrate out-of-order responses when using qemu-io (which is great at sending back-to-back requests prior to waiting for responses) coupled with the nbd file plugin (which has a great feature of rdelay and wdelay, to make it

....errno_is_reliable = plugin_rb_errno_is_reliable, }; NBDKIT_REGISTER_PLUGIN(plugin) diff --git a/src/connections.c b/src/connections.c index cf06ac5..a0d689a 100644 --- a/src/connections.c +++ b/src/connections.c @@ -498,10 +498,7 @@ negotiate_handshake (struct connection *conn) int r; plugin_lock_request (conn); - conn->errno_is_reliable = plugin_errno_is_reliable (conn); - if (conn->errno_is_reliable < 0) - r = -1; - else if (!newstyle) + if (!newstyle) r = _negotiate_handshake_oldstyle (conn); else r = _negotiate_handshake_newstyle (conn); @@ -612,7 +609,7 @@ get_erro...

This fixes the race conditions for me, using the test described here: https://www.redhat.com/archives/libguestfs/2017-September/msg00226.html Rich.

See patch 1 for rationale.

v1 -> v2: - Fixed everything mentioned in the review. Rich.

Somewhat invasive but mostly mechanical change to how plugins are called. This patch is in preparation for adding a second backend subtype for filters. Rich.

On 26.09.2016 14:29, Richard W.M. Jones wrote: > On Mon, Sep 26, 2016 at 02:18:02PM +0200, Carl-Daniel Hailfinger wrote: >> Hi, >> >> has anyone ever run "make check" from nbd against nbdkit with a python >> plugin? I usually get segfaults during such a run, and sometimes various >> other errors happen before the segfault, suggesting that some memory

...+103,7 @@ extern const char *tls_certificates_dir; extern int tls_verify_peer; extern char *unixsocket; extern int verbose; +extern int threads; extern volatile int quit; @@ -151,6 +152,7 @@ extern void plugin_lock_connection (void); extern void plugin_unlock_connection (void); extern void plugin_lock_request (struct connection *conn); extern void plugin_unlock_request (struct connection *conn); +extern bool plugin_is_parallel (void); extern int plugin_errno_is_preserved (void); extern int plugin_open (struct connection *conn, int readonly); extern void plugin_close (struct connection *conn); diff -...

The concern is a client is blocked while processing a request. The nbdkit server design requires a thread per request being processed regardless of the number of connections or clients. We want to run 1000's of requests in parallel without needing a thread at nbdkit layer per request in flight. Our plugin layer is built around boost asio and a few threads in a worker pool running an io

...+ .errno_is_reliable = plugin_rb_errno_is_reliable, }; NBDKIT_REGISTER_PLUGIN(plugin) diff --git a/src/connections.c b/src/connections.c index c0f0567..23e82a9 100644 --- a/src/connections.c +++ b/src/connections.c @@ -498,7 +498,10 @@ negotiate_handshake (struct connection *conn) int r; plugin_lock_request (conn); - if (!newstyle) + conn->errno_is_reliable = plugin_errno_is_reliable (conn); + if (conn->errno_is_reliable < 0) + r = -1; + else if (!newstyle) r = _negotiate_handshake_oldstyle (conn); else r = _negotiate_handshake_newstyle (conn); @@ -602,6 +605,18 @@ validat...

Experimental and only very lightly tested so far. Rich.

The first three patches are identical to: https://www.redhat.com/archives/libguestfs/2018-January/msg00079.html "[PATCH nbdkit v2 0/3] Refactor plugin_* functions into a backend" The rest of the patches add filters using the new filter API previously described here: https://www.redhat.com/archives/libguestfs/2018-January/msg00073.html This needs a lot more testing -- and tests --

This cleans up the existing code base with regards to implicit use of errno from language bindings, then rebases the previous work in python on top of that. I'm still playing with the perl bindings, but got further after reading 'perldoc perlembed'. Eric Blake (4): plugins: Don't use bogus errno from non-C plugins plugins: Add new nbdkit_set_error() utility function python:

A combination of the work that both Rich and I have been doing lately, where filters use only the new API with flags on every command that the client can send over the wire (we can then add support for more flags in nbdkit without having to add new callbacks, as NBD adds more flags upstream). Eric Blake (4): protocol: Split flags from cmd field in requests backend: Pass flags argument through

I've posted some of these patches or ideas before; but now I'm confident enough with the series that it should be ready to push; at any rate, I can now run test-socket-activation in a tight loop without triggering any crashes or hangs. With this in place, I'm going back to work on making the nbd forwarder wort with the parallel thread model. Eric Blake (8): sockets: Use