Re: fast/syscall-free gettimeofday ?
If this was to be extended with cached global syscall information like gettimeofday, would we want that to be in a separate page that is marked non-executable? Is there any way to trick the kernel into leaking arbitrary (and thus executable) code? Also, would it matter for jails? Per-process info like getpid would obviously have to be a separate per-process page. Scott On Jun 14, 2011, at 10:08 AM, K. Macy wrote: http://svnweb.freebsd.org/base/head/sys/sys/imgact.h kib added rudimentary support for this in January On Tue, Jun 14, 2011 at 6:11 PM, Luigi Rizzo ri...@iet.unipi.it wrote: there were discussions at some point on an imprecise but fast implementations of gettimeofday() that would not require a system call (perhaps mmapping some memory region which is opportunistically updated). Does anyone remember what happened about that ? Otherwise, is there any place in the kernel where i can fetch a struct timeval which is not off by more than, say 1 tick ? cheers luigi ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org
Re: fast/syscall-free gettimeofday ?
If this was to be extended with cached global syscall information like gettimeofday, would we want that to be in a separate page that is marked non-executable? Is there any way to trick the kernel into leaking arbitrary (and thus executable) code? Also, would it matter for jails? Per-process info like getpid would obviously have to be a separate per-process page. Scott In the talk about this sort of topic I have seen mention at various times of a page per system, a page per jail, a page per process and a page per thread. I'm not saying we want this all just that I've seen it mentionned.. The per-thread one is the most intersting to do challenge wise. ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org
Re: fast/syscall-free gettimeofday ?
On Jun 15, 2011, at 6:44 PM, Julian Elischer wrote: If this was to be extended with cached global syscall information like gettimeofday, would we want that to be in a separate page that is marked non-executable? Is there any way to trick the kernel into leaking arbitrary (and thus executable) code? Also, would it matter for jails? Per-process info like getpid would obviously have to be a separate per-process page. Scott In the talk about this sort of topic I have seen mention at various times of a page per system, a page per jail, a page per process and a page per thread. I'm not saying we want this all just that I've seen it mentionned.. The per-thread one is the most intersting to do challenge wise. I guess that per-thread would be done via a pointer off of the TLS data, or would it be yet another bumping of the stack? It would be interesting to see how expensive it is to go that direction. Scott ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org
Re: fast/syscall-free gettimeofday ?
On Jun 15, 2011, at 6:44 PM, Julian Elischer wrote: If this was to be extended with cached global syscall information like gettimeofday, would we want that to be in a separate page that is marked non-executable? Is there any way to trick the kernel into leaking arbitrary (and thus executable) code? Also, would it matter for jails? Per-process info like getpid would obviously have to be a separate per-process page. Scott In the talk about this sort of topic I have seen mention at various times of a page per system, a page per jail, a page per process and a page per thread. I'm not saying we want this all just that I've seen it mentionned.. The per-thread one is the most intersting to do challenge wise. I guess that per-thread would be done via a pointer off of the TLS data, or would it be yet another bumping of the stack? It would be interesting to see how expensive it is to go that direction. note that I wasn't volunteering ;-) Scott ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org
Re: fast/syscall-free gettimeofday ?
On Tue, Jun 14, 2011 at 6:08 PM, K. Macy km...@freebsd.org wrote: http://svnweb.freebsd.org/base/head/sys/sys/imgact.h kib added rudimentary support for this in January To clarify, this is just the kernel side of the shared page infrastructure, and is currently only hosting the the signal handling code. The user library side is non-existent and no data sharing currently happens. -Kip On Tue, Jun 14, 2011 at 6:11 PM, Luigi Rizzo ri...@iet.unipi.it wrote: there were discussions at some point on an imprecise but fast implementations of gettimeofday() that would not require a system call (perhaps mmapping some memory region which is opportunistically updated). Does anyone remember what happened about that ? Otherwise, is there any place in the kernel where i can fetch a struct timeval which is not off by more than, say 1 tick ? cheers luigi ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org
Re: fast/syscall-free gettimeofday ?
http://svnweb.freebsd.org/base/head/sys/sys/imgact.h kib added rudimentary support for this in January On Tue, Jun 14, 2011 at 6:11 PM, Luigi Rizzo ri...@iet.unipi.it wrote: there were discussions at some point on an imprecise but fast implementations of gettimeofday() that would not require a system call (perhaps mmapping some memory region which is opportunistically updated). Does anyone remember what happened about that ? Otherwise, is there any place in the kernel where i can fetch a struct timeval which is not off by more than, say 1 tick ? cheers luigi ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to freebsd-current-unsubscr...@freebsd.org