timerfd: Allow timers to be cancelled when clock was set

Some applications must be aware of clock realtime being set
backward. A simple example is a clock applet which arms a timer for
the next minute display. If clock realtime is set backward then the
applet displays a stale time for the amount of time which the clock
was set backwards. Due to that applications poll the time because we
don't have an interface.

Extend the timerfd interface by adding a flag which puts the timer
onto a different internal realtime clock. All timers on this clock are
expired whenever the clock was set.

The timerfd core records the monotonic offset when the timer is
created. When the timer is armed, then the current offset is compared
to the previous recorded offset. When it has changed, then
timerfd_settime returns -ECANCELED. When a timer is read the offset is
compared and if it changed -ECANCELED returned to user space. Periodic
timers are not rearmed in the cancelation case.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <johnstul@us.ibm.com>
Cc: Chris Friesen <chris.friesen@genband.com>
Tested-by: Kay Sievers <kay.sievers@vrfy.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Davide Libenzi <davidel@xmailserver.org>
Reviewed-by: Alexander Shishkin <virtuoso@slind.org>
Link: http://lkml.kernel.org/r/%3Calpine.LFD.2.02.1104271359580.3323%40ionos%3E
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

6 files changed, 111 insertions, 8 deletions

diff --git a/fs/timerfd.c b/fs/timerfd.c
index 8c4fc1425b3..7e14c9e7c4e 100644
--- a/fs/timerfd.c
+++ b/fs/timerfd.c
@@ -26,10 +26,12 @@
 struct timerfd_ctx {
 	struct hrtimer tmr;
 	ktime_t tintv;
+	ktime_t moffs;
 	wait_queue_head_t wqh;
 	u64 ticks;
 	int expired;
 	int clockid;
+	bool might_cancel;
 };
 
 /*
@@ -59,24 +61,52 @@ static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
 	return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
 }
 
-static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
-			  const struct itimerspec *ktmr)
+static bool timerfd_canceled(struct timerfd_ctx *ctx)
+{
+	ktime_t moffs;
+
+	if (!ctx->might_cancel)
+		return false;
+
+	moffs = ktime_get_monotonic_offset();
+
+	if (moffs.tv64 == ctx->moffs.tv64)
+		return false;
+
+	ctx->moffs = moffs;
+	return true;
+}
+
+static int timerfd_setup(struct timerfd_ctx *ctx, int flags,
+			 const struct itimerspec *ktmr)
 {
 	enum hrtimer_mode htmode;
 	ktime_t texp;
+	int clockid = ctx->clockid;
 
 	htmode = (flags & TFD_TIMER_ABSTIME) ?
 		HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
 
+	ctx->might_cancel = false;
+	if (htmode == HRTIMER_MODE_ABS && ctx->clockid == CLOCK_REALTIME &&
+	    (flags & TFD_TIMER_CANCELON_SET)) {
+		clockid = CLOCK_REALTIME_COS;
+		ctx->might_cancel = true;
+	}
+
 	texp = timespec_to_ktime(ktmr->it_value);
 	ctx->expired = 0;
 	ctx->ticks = 0;
 	ctx->tintv = timespec_to_ktime(ktmr->it_interval);
-	hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
+	hrtimer_init(&ctx->tmr, clockid, htmode);
 	hrtimer_set_expires(&ctx->tmr, texp);
 	ctx->tmr.function = timerfd_tmrproc;
-	if (texp.tv64 != 0)
+	if (texp.tv64 != 0) {
 		hrtimer_start(&ctx->tmr, texp, htmode);
+		if (timerfd_canceled(ctx))
+			return -ECANCELED;
+	}
+	return 0;
 }
 
 static int timerfd_release(struct inode *inode, struct file *file)
@@ -118,8 +148,21 @@ static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
 		res = -EAGAIN;
 	else
 		res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);
+
 	if (ctx->ticks) {
 		ticks = ctx->ticks;
+
+		/*
+		 * If clock has changed, we do not care about the
+		 * ticks and we do not rearm the timer. Userspace must
+		 * reevaluate anyway.
+		 */
+		if (timerfd_canceled(ctx)) {
+			ticks = 0;
+			ctx->expired = 0;
+			res = -ECANCELED;
+		}
+
 		if (ctx->expired && ctx->tintv.tv64) {
 			/*
 			 * If tintv.tv64 != 0, this is a periodic timer that
@@ -183,6 +226,7 @@ SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
 	init_waitqueue_head(&ctx->wqh);
 	ctx->clockid = clockid;
 	hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
+	ctx->moffs = ktime_get_monotonic_offset();
 
 	ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
 			       O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
@@ -199,6 +243,7 @@ SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
 	struct file *file;
 	struct timerfd_ctx *ctx;
 	struct itimerspec ktmr, kotmr;
+	int ret;
 
 	if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
 		return -EFAULT;
@@ -240,14 +285,14 @@ SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
 	/*
 	 * Re-program the timer to the new value ...
 	 */
-	timerfd_setup(ctx, flags, &ktmr);
+	ret = timerfd_setup(ctx, flags, &ktmr);
 
 	spin_unlock_irq(&ctx->wqh.lock);
 	fput(file);
 	if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
 		return -EFAULT;
 
-	return 0;
+	return ret;
 }
 
 SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index 4135c88fe4f..eda4ccde073 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -155,6 +155,7 @@ enum  hrtimer_base_type {
 	HRTIMER_BASE_REALTIME,
 	HRTIMER_BASE_MONOTONIC,
 	HRTIMER_BASE_BOOTTIME,
+	HRTIMER_BASE_REALTIME_COS,
 	HRTIMER_MAX_CLOCK_BASES,
 };
 
@@ -310,6 +311,7 @@ extern void hrtimers_resume(void);
 extern ktime_t ktime_get(void);
 extern ktime_t ktime_get_real(void);
 extern ktime_t ktime_get_boottime(void);
+extern ktime_t ktime_get_monotonic_offset(void);
 
 DECLARE_PER_CPU(struct tick_device, tick_cpu_device);
 
diff --git a/include/linux/time.h b/include/linux/time.h
index b3061782dec..a9242773eb2 100644
--- a/include/linux/time.h
+++ b/include/linux/time.h
@@ -302,6 +302,12 @@ struct itimerval {
  * The IDs of various hardware clocks:
  */
 #define CLOCK_SGI_CYCLE			10
+
+#ifdef __KERNEL__
+/* This clock is not exposed to user space */
+#define CLOCK_REALTIME_COS		15
+#endif
+
 #define MAX_CLOCKS			16
 #define CLOCKS_MASK			(CLOCK_REALTIME | CLOCK_MONOTONIC)
 #define CLOCKS_MONO			CLOCK_MONOTONIC
diff --git a/include/linux/timerfd.h b/include/linux/timerfd.h
index 2d0792983f8..e9571fc8f1a 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_TIMER_CANCELON_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_TIMER_CANCELON_SET)
 
 #endif /* _LINUX_TIMERFD_H */
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index c145ed643bc..eabcbd78143 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -78,6 +78,11 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 			.get_time = &ktime_get_boottime,
 			.resolution = KTIME_LOW_RES,
 		},
+		{
+			.index = CLOCK_REALTIME_COS,
+			.get_time = &ktime_get_real,
+			.resolution = KTIME_LOW_RES,
+		},
 	}
 };
 
@@ -85,6 +90,7 @@ static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
 	[CLOCK_REALTIME]	= HRTIMER_BASE_REALTIME,
 	[CLOCK_MONOTONIC]	= HRTIMER_BASE_MONOTONIC,
 	[CLOCK_BOOTTIME]	= HRTIMER_BASE_BOOTTIME,
+	[CLOCK_REALTIME_COS]	= HRTIMER_BASE_REALTIME_COS,
 };
 
 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
@@ -110,6 +116,7 @@ static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 	base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
 	base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
 	base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
+	base->clock_base[HRTIMER_BASE_REALTIME_COS].softirq_time = xtim;
 }
 
 /*
@@ -479,6 +486,8 @@ static inline void debug_deactivate(struct hrtimer *timer)
 	trace_hrtimer_cancel(timer);
 }
 
+static void hrtimer_expire_cancelable(struct hrtimer_cpu_base *cpu_base);
+
 /* High resolution timer related functions */
 #ifdef CONFIG_HIGH_RES_TIMERS
 
@@ -715,9 +724,14 @@ static void retrigger_next_event(void *arg)
 	struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
 	struct timespec realtime_offset, xtim, wtm, sleep;
 
-	if (!hrtimer_hres_active())
+	if (!hrtimer_hres_active()) {
+		raw_spin_lock(&base->lock);
+		hrtimer_expire_cancelable(base);
+		raw_spin_unlock(&base->lock);
 		return;
+	}
 
+	/* Optimized out for !HIGH_RES */
 	get_xtime_and_monotonic_and_sleep_offset(&xtim, &wtm, &sleep);
 	set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec);
 
@@ -727,6 +741,10 @@ static void retrigger_next_event(void *arg)
 		timespec_to_ktime(realtime_offset);
 	base->clock_base[HRTIMER_BASE_BOOTTIME].offset =
 		timespec_to_ktime(sleep);
+	base->clock_base[HRTIMER_BASE_REALTIME_COS].offset =
+		timespec_to_ktime(realtime_offset);
+
+	hrtimer_expire_cancelable(base);
 
 	hrtimer_force_reprogram(base, 0);
 	raw_spin_unlock(&base->lock);
@@ -1222,6 +1240,22 @@ static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
 	timer->state &= ~HRTIMER_STATE_CALLBACK;
 }
 
+static void hrtimer_expire_cancelable(struct hrtimer_cpu_base *cpu_base)
+{
+	struct timerqueue_node *node;
+	struct hrtimer_clock_base *base;
+	ktime_t now = ktime_get_real();
+
+	base = &cpu_base->clock_base[HRTIMER_BASE_REALTIME_COS];
+
+	while ((node = timerqueue_getnext(&base->active))) {
+			struct hrtimer *timer;
+
+			timer = container_of(node, struct hrtimer, node);
+			__run_hrtimer(timer, &now);
+	}
+}
+
 #ifdef CONFIG_HIGH_RES_TIMERS
 
 /*
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index a61b8fa2d39..342408cf68d 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -1099,6 +1099,21 @@ void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
 }
 
 /**
+ * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
+ */
+ktime_t ktime_get_monotonic_offset(void)
+{
+	unsigned long seq;
+	struct timespec wtom;
+
+	do {
+		seq = read_seqbegin(&xtime_lock);
+		wtom = wall_to_monotonic;
+	} while (read_seqretry(&xtime_lock, seq));
+	return timespec_to_ktime(wtom);
+}
+
+/**
  * xtime_update() - advances the timekeeping infrastructure
  * @ticks:	number of ticks, that have elapsed since the last call.
  *