Implement a simple rib entry pending paths queue.
Elements are added to the queue when rib entries are enqueued in
rde_enqueue_updates() for the EVAL_DEFAULT and EVAL_ALL cases.
The queue is then flushed once processed by peer_process_updates().
In rib_remove() assert that the queue is empty before free. This is
implied by the re_is_queued() check but that check only looks at the
pq_mode.
A specific path can only be enqueued once, only the last version matters.
In other words rapid withdraw and update actions will be aggregated into
a single action.
Nothing is currently using this data. This will follow next.
OK tb@
snmpd: convert proc.c to new imsg API (sync with relayd proc.c)
Replace IMSG_SIZE_CHECK() + memcpy() with imsg_get_data(), which does
the length check and copy in one call. Use the imsg accessors
(imsg_get_*) instead of touching imsg.hdr directly and
imsgbuf_get()/imsgbuf_read() instead of imsg_get().
Rewrite proc_forward_imsg() to use imsg_forward() per target imsgbuf
instead of re-composing via proc_compose_imsg(); arm the write event
with imsg_event_add() after each forward.
proc_forward_imsg() never forwarded an fd, and imsg_forward() rewinds
the buffer internally, so multiple forwards per message keep working.
Drop the now-unused n parameter (no callers).
OK rsadowski@
iked: convert proc.c to new imsg API (sync with relayd proc.c)
Replace IMSG_SIZE_CHECK() + memcpy() with imsg_get_data(), which does
the length check and copy in one call. Use the imsg accessors
(imsg_get_*) instead of touching imsg.hdr directly and
imsgbuf_get()/imsgbuf_read() instead of imsg_get().
Rewrite proc_forward_imsg() to use imsg_forward() per target imsgbuf
instead of re-composing via proc_compose_imsg(); arm the write event
with imsg_event_add() after each forward.
proc_forward_imsg() never forwarded an fd, and imsg_forward() rewinds
the buffer internally, so multiple forwards per message keep working.
Drop the now-unused n parameter (all callers passed -1).
OK rsadowski@
httpd: drain abort response via bufferevent
server_dump() did one non-blocking write(2)/tls_write(3) and discarded
the return value, so partial writes were lost, meaning that large
user-defined error documents were silently truncated in transit before
server_close() closed the connection.
OK kirill@
httpd: convert proc.c to new imsg API (sync with relayd proc.c)
Replace IMSG_SIZE_CHECK() + memcpy() with imsg_get_data(), which does
the length check and copy in one call. Use the imsg accessors
(imsg_get_*) instead of touching imsg.hdr directly and
imsgbuf_get()/imsgbuf_read() instead of imsg_get().
Rewrite proc_forward_imsg() to use imsg_forward() per target imsgbuf
instead of re-composing via proc_compose_imsg(); arm the write event
with imsg_event_add() after each forward.
proc_forward_imsg() never forwarded an fd, and imsg_forward() rewinds
the buffer internally, so multiple forwards per message keep working.
Drop the now-unused n parameter (all callers passed -1).
OK martijn@
cleanup when nfsm_srvmtofh1() errors in nfsrv_rename()
corrects vnode references and namei_pool leak
reported by Andrew Griffiths of Calif
initial diff from deraadt@ ok miod@
Pass peer to rde_enqueue_updates() to enqueue updates on.
This allows withdraws to be queued on the peer that caused them and not
on peerself. Also only enqueue a rib entry once and try to keep FIFO
order. Before every call to rde_enqueue_updates() would requeue the
rib entry at the end of the list. Doing this requeue could delay updates
for long time. Further optimisation may need to happen here.
OK tb@
Tighten up the introduction a little:
Mention Match as a conditional directive (previously it only
mentioned Host)
Try to use consistent language in the introduction to refer to
configuration directives (previously it used "parameters" and
"keywords" interchangeably).
Mention that comments may appear at the end of the line too, and that
whitespace at the beginning/end of lines is not significant.
