search for: nbd_concurrent_writer_error

...ither EBADF or worse using another fd that happens to be opened around the same time. -> I think the solution to this would be to allow the writer callback to signal that the socket is about to be closed (eg. add an extra flag parameter to the callback), which would kill the writer thread. (3) nbd_concurrent_writer_error could lose errors. This might happen if the socket is closed normally without writing anything, which would never check h->writer_error. (4) nbd_concurrent_writer_error possibly deadlocks too since it needs to grab h->lock. Basically the same as (1). Rich. -- Richard Jones, Virtualizati...

...iter thread */ > + } > + > +=item 3. Implement writer thread > + > +You must also supply another thread which picks up data off the ring > +and writes it to the socket (see C<nbd_aio_get_fd>). If there an s/there/there is/ > +error when writing to the socket, call C<nbd_concurrent_writer_error> > +with the C<errno>. > + > +You have a choice of whether to implement one thread per nbd_handle or > +one thread shared between all handles. > + > +=item 4. Modify main loop > + > +Finally your main loop can unconditionally call > +C<nbd_aio_notify_write>...

This works, but there's no time saving and I'm still investigating whether it does what I think it does. Nevertheless I thought I would post it because it (probably) implements the idea I had last night outlined in: https://www.redhat.com/archives/libguestfs/2019-June/msg00010.html The meat of the change is patch 4. Patch 5 is an example which I would probably fold into patch 4 for

...t;q_end == NULL) + writer_data->q = writer_data->q_end = item; + else { + writer_data->q_end->next = item; + writer_data->q_end = item; + } + pthread_cond_signal (&writer_data->q_cond); + pthread_mutex_unlock (&writer_data->q_lock); + return; + + error: + nbd_concurrent_writer_error (writer_data->nbd, errno); +} + +static void * +start_writer_thread (void *arg) +{ + struct writer_data *writer_data = arg; + struct nbd_handle *nbd = writer_data->nbd; + struct queue *item; + int fd; + struct pollfd fds[1]; + ssize_t r; + void *p; + + fd = nbd_aio_get_fd (nbd); + +...

v1: https://www.redhat.com/archives/libguestfs/2019-June/msg00014.html I pushed a few bits which are uncontroversial. The main changes since v1 are: An extra patch removes the want_to_send / check for nbd_aio_is_ready in examples/threaded-reads-and-writes.c. This logic was wrong since commit 6af72b87 as was pointed out by Eric in his review. Comments and structure of

...amp;item); + + writer_signal (); /* kick the writer thread */ + } + +=item 3. Implement writer thread + +You must also supply another thread which picks up data off the ring +and writes it to the socket (see C<nbd_aio_get_fd>). If there is an +error when writing to the socket, call C<nbd_concurrent_writer_error> +with the C<errno>. + +You have a choice of whether to implement one thread per nbd_handle or +one thread shared between all handles. + +=item 4. Modify main loop + +Finally your main loop can unconditionally call +C<nbd_aio_notify_write> when C<nbd_aio_get_direction> returns...

...(&item); + + writer_signal (); /* kick the writer thread */ + } + +=item 3. Implement writer thread + +You must also supply another thread which picks up data off the ring +and writes it to the socket (see C<nbd_aio_get_fd>). If there an +error when writing to the socket, call C<nbd_concurrent_writer_error> +with the C<errno>. + +You have a choice of whether to implement one thread per nbd_handle or +one thread shared between all handles. + +=item 4. Modify main loop + +Finally your main loop can unconditionally call +C<nbd_aio_notify_write> when C<nbd_aio_get_direction> returns...

...+ fds[0].events = POLLOUT; + fds[0].revents = 0; + + if (poll (fds, 1, -1) == -1) goto error; + + r = send (fd, p, item->len, 0); + if (r == -1) goto error; + + p += r; + item->len -= r; + } + + free (item->buf); + free (item); + } + + error: + nbd_concurrent_writer_error (nbd, errno); + return NULL; +} -- 2.21.0