On Thu, Apr 18, 2019 at 12:14:58PM -0700, Linus Torvalds wrote:
On Thu, Apr 18, 2019 at 11:58 AM Daniel Colascione dancol@google.com wrote:
On Thu, Apr 18, 2019 at 11:44 AM Jonathan Kowalski bl0pbl33p@gmail.com wrote:
Would using something other than POLLIN be an option (maybe POLLPRI)? The convention is to use it to indicate readability on the descriptor, and also possibly POLLHUP instead of POLLERR (the latter is less of a problem, but FreeBSD also does the same, so it'd help with some consistency for libraries wanting to use this, which aren't interested in other sub states).
Existing event loop libraries generally support checking only for readability and writability. Not setting POLLIN would make these FDs more difficult to use with existing event loop libraries. What advantage would compensate for this difficulty?
Right.
Usually you'd set POLLIN in _addition_ to any other more specialized poll flag.
For example, when a socket has shut down the read side, we do
if (sk->sk_shutdown & RCV_SHUTDOWN) mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;because while it's true that EPOLLRDHUP is the most _specific_ poll bit, it's _also_ true that a read shutdown means that the read() will return immediately.
So generally a HUP condition should mean that POLLIN and POLLOUT also get set. Not because there's any actual _data_ to be read, but simply because the read will not block.
Sounds great and I agree with Linus and Daniel. So I am guessing you are Ok with the current set of flags proposed this patch, so I will keep them intact in future patch postings. But please let me know if you want me to change something about the flags.
thanks!
- Joel