Re: [RFCv4] timerfd: add TFD_NOTIFY_CLOCK_SET to watch for clockchanges

From: Kirill A. Shutemov
Date: Thu Mar 10 2011 - 03:03:00 EST

On Wed, Mar 09, 2011 at 04:36:51PM +0200, Alexander Shishkin wrote:
> Changes since v3:
> - changed timerfd_settime() semantics (see below)
> Changes since v2:
> - replaced sysfs interface with a syscall
> - added sysctl/procfs handle to set a limit to the number of users
> - fixed issues pointed out by Greg.
> Changes since v1:
> - updated against 2.6.36-rc1,
> - added notification/filtering options,
> - added Documentation/ABI/sysfs-kernel-time-notify interface description.
>
> Certain userspace applications (like "clock" desktop applets or cron or
> systemd) might want to be notified when some other application changes
> the system time. There are several known to me reasons for this:
> - avoiding periodic wakeups to poll time changes;
> - rearming CLOCK_REALTIME timers when said changes happen;
> - changing system timekeeping policy for system-wide time management
> programs;
> - keeping guest applications/operating systems running in emulators
> up to date;
> - recalculation of traffic signal cycles in Advanced Traffic Controllers
> (ATC), which is part of ATC API requirements [1] as developed by ATC
> working group of the U.S. Institute of Transportation Engineers (ITE).
>
> The major change since the previous version is the new semantics of
> timerfd_settime() when it's called on a time change notification
> descriptor: it will set the system time to utmr.it_value if the time
> change counter is zero, otherwise it will return EBUSY, this is required
> to prevent a race between setting the time and reading the counter, when
> the time controlling procees changes the time immediately after another
> process in the system did the same (the counter is greater than one),
> that process' time change will be lost. Thus, the time controlling
> process should use timerfd_settime() instead of clock_settime() or
> settimeofday() to ensure that other processes' time changes don't get
> lost.
>
> This is another attempt to approach notifying userspace about system
> clock changes. The other one is using an eventfd and a syscall [1]. In
> the course of discussing the necessity of a syscall for this kind of
> notifications, it was suggested that this functionality can be achieved
> via timers [2] (and timerfd in particular [3]). This idea got quite
> some support [4], [5], [6] and some vague criticism [7], so I decided
> to try and go a bit further with it.
>
> [1] http://www.ite.org/standards/atcapi/version2.asp
> [2] http://marc.info/?l=linux-kernel&m=128950389423614&w=2
> [3] http://marc.info/?l=linux-kernel&m=128951020831573&w=2
> [4] http://marc.info/?l=linux-kernel&m=128951588006157&w=2
> [5] http://marc.info/?l=linux-kernel&m=128951503205111&w=2
> [6] http://marc.info/?l=linux-kernel&m=128955890118477&w=2
> [7] http://marc.info/?l=linux-kernel&m=129002967031104&w=2
> [8] http://marc.info/?l=linux-kernel&m=129002672227263&w=2
>
> Signed-off-by: Alexander Shishkin <virtuoso@xxxxxxxxx>
> CC: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> CC: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>
> CC: Greg Kroah-Hartman <gregkh@xxxxxxx>
> CC: Feng Tang <feng.tang@xxxxxxxxx>
> CC: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> CC: Michael Tokarev <mjt@xxxxxxxxxx>
> CC: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
> CC: John Stultz <johnstul@xxxxxxxxxx>
> CC: Chris Friesen <chris.friesen@xxxxxxxxxxx>
> CC: Kay Sievers <kay.sievers@xxxxxxxx>
> CC: Kirill A. Shutemov <kirill@xxxxxxxxxxxxx>
> CC: Artem Bityutskiy <dedekind1@xxxxxxxxx>
> CC: Davide Libenzi <davidel@xxxxxxxxxxxxxxx>
> CC: linux-fsdevel@xxxxxxxxxxxxxxx
> CC: linux-kernel@xxxxxxxxxxxxxxx
> ---
> fs/timerfd.c | 94 ++++++++++++++++++++++++++++++++++++++++++-----
> include/linux/hrtimer.h | 6 +++
> include/linux/timerfd.h | 3 +-
> kernel/compat.c | 5 ++-
> kernel/hrtimer.c | 4 ++
> kernel/time.c | 11 ++++-
> 6 files changed, 109 insertions(+), 14 deletions(-)
>
> diff --git a/fs/timerfd.c b/fs/timerfd.c
> index 8c4fc14..6170f61 100644
> --- a/fs/timerfd.c
> +++ b/fs/timerfd.c
> @@ -22,6 +22,7 @@
> #include <linux/anon_inodes.h>
> #include <linux/timerfd.h>
> #include <linux/syscalls.h>
> +#include <linux/security.h>
>
> struct timerfd_ctx {
> struct hrtimer tmr;
> @@ -30,8 +31,13 @@ struct timerfd_ctx {
> u64 ticks;
> int expired;
> int clockid;
> + struct list_head notifiers_list;
> };
>
> +/* TFD_NOTIFY_CLOCK_SET timers go here */
> +static DEFINE_SPINLOCK(notifiers_lock);
> +static LIST_HEAD(notifiers_list);
> +
> /*
> * This gets called when the timer event triggers. We set the "expired"
> * flag, but we do not re-arm the timer (in case it's necessary,
> @@ -51,10 +57,31 @@ static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
> return HRTIMER_NORESTART;
> }
>
> +void timerfd_clock_was_set(clockid_t clockid)
> +{
> + struct timerfd_ctx *ctx;
> + unsigned long flags;
> +
> + spin_lock(&notifiers_lock);
> + list_for_each_entry(ctx, &notifiers_list, notifiers_list) {
> + spin_lock_irqsave(&ctx->wqh.lock, flags);
> + if (ctx->tmr.base->index == clockid) {
> + ctx->ticks++;
> + wake_up_locked(&ctx->wqh);
> + }
> + spin_unlock_irqrestore(&ctx->wqh.lock, flags);
> + }
> + spin_unlock(&notifiers_lock);
> +}
> +
> static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
> {
> ktime_t remaining;
>
> + /* for notification timers, return current time */
> + if (!list_empty(&ctx->notifiers_list))
> + return timespec_to_ktime(current_kernel_time());
> +
> remaining = hrtimer_expires_remaining(&ctx->tmr);
> return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
> }
> @@ -72,6 +99,12 @@ static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
> ctx->expired = 0;
> ctx->ticks = 0;
> ctx->tintv = timespec_to_ktime(ktmr->it_interval);
> +
> + if (flags & TFD_NOTIFY_CLOCK_SET) {
> + list_add(&ctx->notifiers_list, &notifiers_list);
> + return;
> + }
> +
> hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
> hrtimer_set_expires(&ctx->tmr, texp);
> ctx->tmr.function = timerfd_tmrproc;
> @@ -83,7 +116,12 @@ static int timerfd_release(struct inode *inode, struct file *file)
> {
> struct timerfd_ctx *ctx = file->private_data;
>
> - hrtimer_cancel(&ctx->tmr);
> + if (!list_empty(&ctx->notifiers_list)) {
> + spin_lock(&notifiers_lock);
> + list_del(&ctx->notifiers_list);
> + spin_unlock(&notifiers_lock);
> + } else
> + hrtimer_cancel(&ctx->tmr);
> kfree(ctx);
> return 0;
> }
> @@ -113,6 +151,7 @@ static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
>
> if (count < sizeof(ticks))
> return -EINVAL;
> +
> spin_lock_irq(&ctx->wqh.lock);
> if (file->f_flags & O_NONBLOCK)
> res = -EAGAIN;

Whitespace changes?

> @@ -120,7 +159,8 @@ static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
> res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);
> if (ctx->ticks) {
> ticks = ctx->ticks;
> - if (ctx->expired && ctx->tintv.tv64) {
> + if (ctx->expired && ctx->tintv.tv64 &&
> + list_empty(&ctx->notifiers_list)) {
> /*
> * If tintv.tv64 != 0, this is a periodic timer that
> * needs to be re-armed. We avoid doing it in the timer
> @@ -184,6 +224,8 @@ SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
> ctx->clockid = clockid;
> hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
>
> + INIT_LIST_HEAD(&ctx->notifiers_list);
> +
> ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
> O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
> if (ufd < 0)
> @@ -196,18 +238,24 @@ SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
> const struct itimerspec __user *, utmr,
> struct itimerspec __user *, otmr)
> {
> + int ret = 0;
> struct file *file;
> struct timerfd_ctx *ctx;
> struct itimerspec ktmr, kotmr;
>
> - if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
> - return -EFAULT;
> -
> - if ((flags & ~TFD_SETTIME_FLAGS) ||
> - !timespec_valid(&ktmr.it_value) ||
> - !timespec_valid(&ktmr.it_interval))
> + if (flags & ~TFD_SETTIME_FLAGS)
> return -EINVAL;
>
> + /* utmr may be NULL for notification timerfd */
> + if (!(flags & TFD_NOTIFY_CLOCK_SET) || utmr) {
> + if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
> + return -EFAULT;
> +
> + if (!timespec_valid(&ktmr.it_value) ||
> + !timespec_valid(&ktmr.it_interval))
> + return -EINVAL;
> + }
> +
> file = timerfd_fget(ufd);
> if (IS_ERR(file))
> return PTR_ERR(file);
> @@ -218,10 +266,12 @@ SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
> * it to the new values.
> */
> for (;;) {
> + spin_lock(&notifiers_lock);
> spin_lock_irq(&ctx->wqh.lock);
> - if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
> + if (!list_empty(&notifiers_list) || hrtimer_try_to_cancel(&ctx->tmr) >= 0)
> break;
> spin_unlock_irq(&ctx->wqh.lock);
> + spin_unlock(&notifiers_lock);
> cpu_relax();
> }
>
> @@ -238,16 +288,39 @@ SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
> kotmr.it_interval = ktime_to_timespec(ctx->tintv);
>
> /*
> + * for the notification timerfd, set current time to it_value
> + * if the timer hasn't expired; otherwise someone has changed
> + * the system time to the value that we don't know
> + */
> + if (!list_empty(&ctx->notifiers_list) && utmr) {
> + if (ctx->ticks) {
> + ret = -EBUSY;
> + goto out;
> + }
> +
> + ret = security_settime(&ktmr.it_value, NULL);
> + if (ret)
> + goto out;
> +
> + spin_unlock_irq(&ctx->wqh.lock);
> + ret = do_settimeofday(&ktmr.it_value);
> + goto out1;
> + }
> +
> + /*
> * Re-program the timer to the new value ...
> */
> timerfd_setup(ctx, flags, &ktmr);
>
> +out:
> spin_unlock_irq(&ctx->wqh.lock);
> +out1:
> + spin_unlock(&notifiers_lock);
> fput(file);
> if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
> return -EFAULT;

What's reason to do copy_to_user() in case of error?
Is it safe for error from security_settime()?

> - return 0;
> + return ret;
> }
>
> SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
> @@ -273,6 +346,7 @@ SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
> spin_unlock_irq(&ctx->wqh.lock);
> fput(file);
>
> +out:
> return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
> }

This lable is not needed.

> diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
> index 6bc1804..991e8d9 100644
> --- a/include/linux/hrtimer.h
> +++ b/include/linux/hrtimer.h
> @@ -248,6 +248,12 @@ static inline ktime_t hrtimer_expires_remaining(const struct hrtimer *timer)
> return ktime_sub(timer->node.expires, timer->base->get_time());
> }
>
> +#ifdef CONFIG_TIMERFD
> +extern void timerfd_clock_was_set(clockid_t clockid);
> +#else
> +static inline void timerfd_clock_was_set(clockid_t clockid) {}
> +#endif
> +
> #ifdef CONFIG_HIGH_RES_TIMERS
> struct clock_event_device;
>
> diff --git a/include/linux/timerfd.h b/include/linux/timerfd.h
> index 2d07929..c3ddad9 100644
> --- a/include/linux/timerfd.h
> +++ b/include/linux/timerfd.h
> @@ -19,6 +19,7 @@
> * shared O_* flags.
> */
> #define TFD_TIMER_ABSTIME (1 << 0)
> +#define TFD_NOTIFY_CLOCK_SET (1 << 1)
> #define TFD_CLOEXEC O_CLOEXEC
> #define TFD_NONBLOCK O_NONBLOCK
>
> @@ -26,6 +27,6 @@
> /* Flags for timerfd_create. */
> #define TFD_CREATE_FLAGS TFD_SHARED_FCNTL_FLAGS
> /* Flags for timerfd_settime. */
> -#define TFD_SETTIME_FLAGS TFD_TIMER_ABSTIME
> +#define TFD_SETTIME_FLAGS (TFD_TIMER_ABSTIME | TFD_NOTIFY_CLOCK_SET)
>
> #endif /* _LINUX_TIMERFD_H */
> diff --git a/kernel/compat.c b/kernel/compat.c
> index 38b1d2c..b1cf3e1 100644
> --- a/kernel/compat.c
> +++ b/kernel/compat.c
> @@ -995,7 +995,10 @@ asmlinkage long compat_sys_stime(compat_time_t __user *tptr)
> if (err)
> return err;
>
> - do_settimeofday(&tv);
> + err = do_settimeofday(&tv);
> + if (!err)
> + timerfd_clock_was_set(CLOCK_REALTIME);
> +
> return 0;
> }
>
> diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
> index 2c3d6e5..469eef6 100644
> --- a/kernel/hrtimer.c
> +++ b/kernel/hrtimer.c
> @@ -663,6 +663,7 @@ void clock_was_set(void)
> {
> /* Retrigger the CPU local events everywhere */
> on_each_cpu(retrigger_next_event, NULL, 1);
> +
> }
>
> /*

Whitespace changes?

> @@ -675,6 +676,9 @@ void hres_timers_resume(void)
> KERN_INFO "hres_timers_resume() called with IRQs enabled!");
>
> retrigger_next_event(NULL);
> +
> + /* Trigger timerfd notifiers */
> + timerfd_clock_was_set(CLOCK_MONOTONIC);
> }
>
> /*
> diff --git a/kernel/time.c b/kernel/time.c
> index 6430a75..b06f759 100644
> --- a/kernel/time.c
> +++ b/kernel/time.c
> @@ -92,7 +92,10 @@ SYSCALL_DEFINE1(stime, time_t __user *, tptr)
> if (err)
> return err;
>
> - do_settimeofday(&tv);
> + err = do_settimeofday(&tv);
> + if (!err)
> + timerfd_clock_was_set(CLOCK_REALTIME);
> +
> return 0;
> }
>
> @@ -177,7 +180,11 @@ int do_sys_settimeofday(const struct timespec *tv, const struct timezone *tz)
> /* SMP safe, again the code in arch/foo/time.c should
> * globally block out interrupts when it runs.
> */
> - return do_settimeofday(tv);
> + error = do_settimeofday(tv);
> + if (!error)
> + timerfd_clock_was_set(CLOCK_REALTIME);
> +
> + return error;
> }
> return 0;
> }
> --
> 1.7.2.1.45.gb66c2
>

--
Kirill A. Shutemov
--
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/