Age | Commit message (Collapse) | Author |
|
|
|
isn't anything left to be written. Signal imsgev termination by
setting iev->terminate = 1(Mind that we are setting it from with
the callback). As imsgev would again callback dispatch with
IMSGEV_DONE, we cannot free imsgev yet. So iev_maildrop need to
exist beyond lifetime of session until IMSGEV_DONE. Allocating it
seperately from session.
imsgev_close is avoided as it schedules another EV_WRITE which is
not needed in our case.
This fixes a crash observed consistently with a little perl script
to simulate concurrent sessions. Thanks MALLOC_OPTIONS='SFG<<'
|
|
|
|
|
|
process of MAILDROP_INIT. Maildrop now sets up itself and waits for
MAILDROP_INIT from session.
This would now eliminate the doubt in the earlier setup when priv
process forks and the child gets cpu slice first before parent which
sends MAILDROP_INIT to an uninitialized imsgev.
|
|
|
|
|
|
|
|
Turner (#5).
imsgev_close schedules(event_add) iev and the event struct is free-ed
before the event is dispatched.
Include string.h after recent libressl changes.
|
|
|
|
|
|
check required here.
|
|
session.
|
|
|
|
consolidating into one imsg. We are no longer limited by MAX_IMSGSIZE.
End of list is signalled by sending an imsg with hash as "" or sz
as 0.
|
|
|