* Miklos Szeredi [off-list ref] 2007-06-18 11:44
quoted
Garbage collection only ever happens, if the app is sending AF_UNIX
sockets over AF_UNIX sockets. Which is a rather rare case. And which
is basically why this bug went unnoticed for so long.
So my second patch only affects the performance of _exactly_ those
apps which might well be bitten by the bug itself.
That's not entirely the truth. It affects all applications using
AF_UNIX sockets while file descriptors are being transfered. I
agree that the performance impact is not severe on most systems
but if file descriptors are being transfered continously by just
a single application it can become rather severe.
You are wrong. Look in unix_release_sock():
if (atomic_read(&unix_tot_inflight))
unix_gc(); /* Garbage collect fds */
unix_tot_inflight is the number of AF_UNIX sockets currently being
transferred over some AF_UNIX sockets.
That means that just sending (non-unix socket) fds over unix sockets
will never invoke the gc.
Miklos
* Miklos Szeredi [off-list ref] 2007-06-18 12:39
You are wrong. Look in unix_release_sock():
if (atomic_read(&unix_tot_inflight))
unix_gc(); /* Garbage collect fds */
unix_tot_inflight is the number of AF_UNIX sockets currently being
transferred over some AF_UNIX sockets.
That means that just sending (non-unix socket) fds over unix sockets
will never invoke the gc.
That's what I meant, I'm sorry, I should have written unix socket
file descriptor to not leave any room for misinterpretation.
On Mon, 18 Jun 2007 12:43:40 +0200
Thomas Graf [off-list ref] wrote:
* Miklos Szeredi [off-list ref] 2007-06-18 12:39
quoted
You are wrong. Look in unix_release_sock():
if (atomic_read(&unix_tot_inflight))
unix_gc(); /* Garbage collect fds */
unix_tot_inflight is the number of AF_UNIX sockets currently being
transferred over some AF_UNIX sockets.
That means that just sending (non-unix socket) fds over unix sockets
will never invoke the gc.
That's what I meant, I'm sorry, I should have written unix socket
file descriptor to not leave any room for misinterpretation.
You can bound the worst case on this and I think stay within the specs
(as the specs don't say a lot about it). One way would be to make
unix_gc() kick off a thread/tasklet and if the last unix_gc was within 5
seconds then use a timer to defer it - that prevents any user driven
"lets cause a ton of gc" cases.