Re: [PATCH] [RFCv2] notify userspace about time changes

[Greg KH]

On Wed, Aug 18, 2010 at 04:55:39PM +0300, Alexander Shishkin wrote:
> --- /dev/null
> +++ b/Documentation/ABI/testing/sysfs-kernel-time-notify
> @@ -0,0 +1,39 @@
> +What:		/sys/kernel/time_notify
> +Date:		August 2010
> +Contact:	Alexander Shishkin <virtuoso@xxxxxxxxx>
> +Description:
> +		This file is used by processes which want to receive
> +		notifications about system time changes, to communicate their
> +		eventfd descriptor number along with the type of event they
> +		wish to subscribe to and filtering options. Every time one of
> +		the processes in the system succeeds in doing a settimeofday()
> +		or adjtimex(), this eventfd will be signalled.
> +		This file is not readable. Upon write, this is the format
> +		accepted by the kernel:
> +		<fd> <want-others> <want-own> <want-set> <want-adj>
> +		where
> +		- <fd> is the file descriptor of the eventfd that the process
> +		will listen to for time change events;
> +		- <want-others> is 0 or 1 depending on whether the process wants
> +		to be notified about other processes' time changes;
> +		- <want-own> -- likewise, for process' own time changes;
> +		- <want-set> is 0 or 1 depending on whether the process wants
> +		to be notified about settimeofday()/stime() system calls;
> +		- <want-adj> -- likewise, for adjtimex() system call.
> +
> +		A simple snippet of C code illustrates the usage pattern:
> +
> +		     efd = eventfd(0, 0);
> +		     fd = open("/sys/kernel/time_notify", O_WRONLY);
> +		     fdprintf(fd, "%d 1 0 1 1", efd);
> +		     close(fd);
> +
> +		     fds[0].fd = efd;
> +		     fds[0].events = POLLIN;
> +		     while (poll(fds, 1, -1) > 0) {
> +		           uint64_t n;
> +
> +			   read(efd, &n, 8);
> +			   printf("time has been modified %d time(s)\n", n);
> +		     }

While this isn't the "nicest" sysfs file ever, it's not all that bad.


> +/* time change events */
> +#define TIME_EVENT_SET 0
> +#define TIME_EVENT_ADJ 1
> +
> +#ifdef CONFIG_TIME_NOTIFY
> +void time_notify_all(int type);
> +#else
> +#define time_notify_all(x) do {} while (0)

static inline please so we do a typecheck.

> +#endif
> +
>  #endif /* __KERNEL__ */
>  
>  #define NFDBITS			__NFDBITS
> diff --git a/init/Kconfig b/init/Kconfig
> index 2de5b1c..aa4d890 100644
> --- a/init/Kconfig
> +++ b/init/Kconfig
> @@ -980,6 +980,14 @@ config PERF_USE_VMALLOC
>  	help
>  	  See tools/perf/design.txt for details
>  
> +config TIME_NOTIFY
> +	bool
> +	default y

Really?  You obviously haven't added new Kconfig options to the kernel
in the past and been yelled at by Linus :)

> +	evt->eventfd = eventfd_ctx_fileget(file);
> +	if (!evt->eventfd) {
> +		ret = PTR_ERR(evt->eventfd);

Are you sure this is correct?

thanks,

greg k-h
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/