[patch V2 00/17] posix-timers: Rework the global hash table and provide a sane mechanism for CRIU

[patch V2 00/17] posix-timers: Rework the global hash table and provide a sane mechanism for CRIU

[Thomas Gleixner]

This is a follow up on V1 of this work, which can be found here:

     https://lore.kernel.org/all/20250224095736.145530367@linutronix.de

It addresses the scalability problem of the posix timer hash and provides a
performant mechanism to restore Posix timers with a given ID along with a
couple of preperatory cleanups and enhancements. More details about
implementation choices are in the change logs and the cover letter of V1.

Changes vs. V1:

  - Simplify the timer removal cleanup - PeterZ
  - Use guards consistently - PeterZ
  - Switch to jhash32()
  - Avoid false sharing by reordering struct k_itimer
  - Advance the ID counter when restoring
  - A few additional cleanups

The series survives all posix timer tests and did not show any regressions
so far.

The series is based on:

    git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip timers/core

and is also available from git:

    git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git timers/posix

Thanks,

	tglx
---
Eric Dumazet (3):
      posix-timers: Initialise timer before adding it to the hash table
      posix-timers: Add cond_resched() to posix_timer_add() search loop
      posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

Peter Zijlstra (1):
      posix-timers: Make lock_timer() use guard()

Thomas Gleixner (13):
      posix-timers: Cleanup includes
      posix-timers: Remove a few paranoid warnings
      posix-timers: Remove SLAB_PANIC from kmem cache
      posix-timers: Use guards in a few places
      posix-timers: Simplify lock/unlock_timer()
      posix-timers: Rework timer removal
      posix-timers: Improve hash table performance
      posix-timers: Switch to jhash32()
      posix-timers: Avoid false cacheline sharing
      posix-timers: Make per process list RCU safe
      posix-timers: Don't iterate /proc/$PID/timers with sighand::siglock held
      posix-timers: Provide a mechanism to allocate a given timer ID
      selftests/timers/posix-timers: Add a test for exact allocation mode

 fs/proc/base.c                                |   48 --
 include/linux/cleanup.h                       |   22 -
 include/linux/posix-timers.h                  |   30 +
 include/linux/sched/signal.h                  |    3 
 include/uapi/linux/prctl.h                    |   10 
 kernel/signal.c                               |    2 
 kernel/sys.c                                  |    5 
 kernel/time/posix-timers.c                    |  530 +++++++++++++-------------
 tools/testing/selftests/timers/posix_timers.c |   60 ++
 9 files changed, 403 insertions(+), 307 deletions(-)

[patch V2 01/17] posix-timers: Initialise timer before adding it to the hash table

[Thomas Gleixner]

From: Eric Dumazet <edumazet@google.com>

A timer is only valid in the hashtable when both timer::it_signal and
timer::it_id are set to their final values, but timers are added without
those values being set.

The timer ID is allocated when the timer is added to the hash in invalid
state. The ID is taken from a monotonically increasing per process counter
which wraps around after reaching INT_MAX. The hash insertion validates
that there is no timer with the allocated ID in the hash table which
belongs to the same process. That opens a mostly theoretical race condition:

If other threads of the same process manage to create/delete timers in
rapid succession before the newly created timer is fully initialized and
wrap around to the timer ID which was handed out, then a duplicate timer ID
will be inserted into the hash table.

Prevent this by:

  1) Setting timer::it_id before inserting the timer into the hashtable.
 
  2) Storing the signal pointer in timer::it_signal with bit 0 set before
     inserting it into the hashtable.

     Bit 0 acts as a invalid bit, which means that the regular lookup for
     sys_timer_*() will fail the comparison with the signal pointer.

     But the lookup on insertion masks out bit 0 and can therefore detect a
     timer which is not yet valid, but allocated in the hash table.  Bit 0
     in the pointer is cleared once the initialization of the timer
     completed.

[ tglx: Fold ID and signal iniitializaion into one patch and massage change
  	log and comments. ]

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250219125522.2535263-3-edumazet@google.com

---
 kernel/time/posix-timers.c |   56 +++++++++++++++++++++++++++++++++------------
 1 file changed, 42 insertions(+), 14 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -72,13 +72,13 @@ static int hash(struct signal_struct *si
 	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
 }
 
-static struct k_itimer *__posix_timers_find(struct hlist_head *head,
-					    struct signal_struct *sig,
-					    timer_t id)
+static struct k_itimer *posix_timer_by_id(timer_t id)
 {
+	struct signal_struct *sig = current->signal;
+	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
 	struct k_itimer *timer;
 
-	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
+	hlist_for_each_entry_rcu(timer, head, t_hash) {
 		/* timer->it_signal can be set concurrently */
 		if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
 			return timer;
@@ -86,12 +86,26 @@ static struct k_itimer *__posix_timers_f
 	return NULL;
 }
 
-static struct k_itimer *posix_timer_by_id(timer_t id)
+static inline struct signal_struct *posix_sig_owner(const struct k_itimer *timer)
 {
-	struct signal_struct *sig = current->signal;
-	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
+	unsigned long val = (unsigned long)timer->it_signal;
+
+	/*
+	 * Mask out bit 0, which acts as invalid marker to prevent
+	 * posix_timer_by_id() detecting it as valid.
+	 */
+	return (struct signal_struct *)(val & ~1UL);
+}
+
+static bool posix_timer_hashed(struct hlist_head *head, struct signal_struct *sig, timer_t id)
+{
+	struct k_itimer *timer;
 
-	return __posix_timers_find(head, sig, id);
+	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
+		if ((posix_sig_owner(timer) == sig) && (timer->it_id == id))
+			return true;
+	}
+	return false;
 }
 
 static int posix_timer_add(struct k_itimer *timer)
@@ -112,7 +126,19 @@ static int posix_timer_add(struct k_itim
 		sig->next_posix_timer_id = (id + 1) & INT_MAX;
 
 		head = &posix_timers_hashtable[hash(sig, id)];
-		if (!__posix_timers_find(head, sig, id)) {
+		if (!posix_timer_hashed(head, sig, id)) {
+			/*
+			 * Set the timer ID and the signal pointer to make
+			 * it identifiable in the hash table. The signal
+			 * pointer has bit 0 set to indicate that it is not
+			 * yet fully initialized. posix_timer_hashed()
+			 * masks this bit out, but the syscall lookup fails
+			 * to match due to it being set. This guarantees
+			 * that there can't be duplicate timer IDs handed
+			 * out.
+			 */
+			timer->it_id = (timer_t)id;
+			timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
 			hlist_add_head_rcu(&timer->t_hash, head);
 			spin_unlock(&hash_lock);
 			return id;
@@ -406,8 +432,7 @@ static int do_timer_create(clockid_t whi
 
 	/*
 	 * Add the timer to the hash table. The timer is not yet valid
-	 * because new_timer::it_signal is still NULL. The timer id is also
-	 * not yet visible to user space.
+	 * after insertion, but has a unique ID allocated.
 	 */
 	new_timer_id = posix_timer_add(new_timer);
 	if (new_timer_id < 0) {
@@ -415,7 +440,6 @@ static int do_timer_create(clockid_t whi
 		return new_timer_id;
 	}
 
-	new_timer->it_id = (timer_t) new_timer_id;
 	new_timer->it_clock = which_clock;
 	new_timer->kclock = kc;
 	new_timer->it_overrun = -1LL;
@@ -453,7 +477,7 @@ static int do_timer_create(clockid_t whi
 	}
 	/*
 	 * After succesful copy out, the timer ID is visible to user space
-	 * now but not yet valid because new_timer::signal is still NULL.
+	 * now but not yet valid because new_timer::signal low order bit is 1.
 	 *
 	 * Complete the initialization with the clock specific create
 	 * callback.
@@ -463,7 +487,11 @@ static int do_timer_create(clockid_t whi
 		goto out;
 
 	spin_lock_irq(&current->sighand->siglock);
-	/* This makes the timer valid in the hash table */
+	/*
+	 * new_timer::it_signal contains the signal pointer with bit 0 set,
+	 * which makes it invalid for syscall operations. Store the
+	 * unmodified signal pointer to make it valid.
+	 */
 	WRITE_ONCE(new_timer->it_signal, current->signal);
 	hlist_add_head(&new_timer->list, &current->signal->posix_timers);
 	spin_unlock_irq(&current->sighand->siglock);

Re: [patch V2 01/17] posix-timers: Initialise timer before adding it to the hash table

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:44PM +0100, Thomas Gleixner a écrit :
> From: Eric Dumazet <edumazet@google.com>
> 
> A timer is only valid in the hashtable when both timer::it_signal and
> timer::it_id are set to their final values, but timers are added without
> those values being set.
> 
> The timer ID is allocated when the timer is added to the hash in invalid
> state. The ID is taken from a monotonically increasing per process counter
> which wraps around after reaching INT_MAX. The hash insertion validates
> that there is no timer with the allocated ID in the hash table which
> belongs to the same process. That opens a mostly theoretical race condition:
> 
> If other threads of the same process manage to create/delete timers in
> rapid succession before the newly created timer is fully initialized and
> wrap around to the timer ID which was handed out, then a duplicate timer ID
> will be inserted into the hash table.
> 
> Prevent this by:
> 
>   1) Setting timer::it_id before inserting the timer into the hashtable.
>  
>   2) Storing the signal pointer in timer::it_signal with bit 0 set before
>      inserting it into the hashtable.
> 
>      Bit 0 acts as a invalid bit, which means that the regular lookup for
>      sys_timer_*() will fail the comparison with the signal pointer.
> 
>      But the lookup on insertion masks out bit 0 and can therefore detect a
>      timer which is not yet valid, but allocated in the hash table.  Bit 0
>      in the pointer is cleared once the initialization of the timer
>      completed.
> 
> [ tglx: Fold ID and signal iniitializaion into one patch and massage change
>   	log and comments. ]
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Link: https://lore.kernel.org/all/20250219125522.2535263-3-edumazet@google.com

Looking at this more or less lockless whole thing again, is the
ordering between creation and subsequent operations sufficiently guaranteed?

    T0                                                T1
---------                                             -----------
do_timer_create()
    posix_timer_add()
        spin_lock(hash_lock)
        // A
        timer->it_id = ...
        spin_unlock(hash_lock)
    // Initialize timer fields
    // B
    new_timer->.... = ....
    common_timer_create()
        // C
        hrtimer_init()
    spin_lock(current->sighand)
    // D
    WRITE_ONCE(new_timer->it_signal, current->signal)
    spin_unlock(current->sighand)
                                                      do_timer_settime()
                                                          lock_timer()
                                                              // observes A && D
                                                              posix_timer_by_id()
                                                              spin_lock_irqsave(&timr->it_lock)
                                                              // recheck ok
                                                              if (timr->it_signal == current->signal)
                                                                  return timr
                                                              common_timer_get()
                                                                  // fiddle with timer fields
                                                                  // but doesn't observe B
                                                                  // for example doesn't observe SIGEV_NONE
                                                                  sig_none = timr->it_sigev_notify == SIGEV_NONE;
                                                                  ...
                                                                  // doesn't observe C
                                                                  // hrtimer_init() isn't visible yet
                                                                  // It might mess up after the hrtimer_start()
                                                                  hrtimer_start()

How about the following instead?

diff --git a/kernel/time/posix-timers.c b/kernel/time/posix-timers.c
index 44ba7db07e90..8769a1ccf69a 100644
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -72,26 +72,29 @@ static int hash(struct signal_struct *sig, unsigned int nr)
 	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
 }
 
-static struct k_itimer *__posix_timers_find(struct hlist_head *head,
-					    struct signal_struct *sig,
-					    timer_t id)
+static struct k_itimer *posix_timer_by_id(timer_t id)
 {
+	struct signal_struct *sig = current->signal;
+	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
 	struct k_itimer *timer;
 
-	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
-		/* timer->it_signal can be set concurrently */
-		if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
+	hlist_for_each_entry_rcu(timer, head, t_hash) {
+		if (timer->it_signal == sig && timer->it_id == id &&
+		    !hlist_unhashed_lockless(&timer->list))
 			return timer;
 	}
 	return NULL;
 }
 
-static struct k_itimer *posix_timer_by_id(timer_t id)
+static bool posix_timer_hashed(struct hlist_head *head, struct signal_struct *sig, timer_t id)
 {
-	struct signal_struct *sig = current->signal;
-	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
+	struct k_itimer *timer;
 
-	return __posix_timers_find(head, sig, id);
+	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
+		if (timer->it_signal == sig && timer->it_id == id)
+			return true;
+	}
+	return false;
 }
 
 static int posix_timer_add(struct k_itimer *timer)
@@ -112,7 +115,15 @@ static int posix_timer_add(struct k_itimer *timer)
 		sig->next_posix_timer_id = (id + 1) & INT_MAX;
 
 		head = &posix_timers_hashtable[hash(sig, id)];
-		if (!__posix_timers_find(head, sig, id)) {
+		if (!posix_timer_hashed(head, sig, id)) {
+			/*
+			 * Set the timer ID and the signal pointer to make
+			 * it identifiable in the global hash table. Its
+			 * lookup is only valid to lock_timer() once it is
+			 * hashed within the process itself.
+			 */
+			timer->it_id = (timer_t)id;
+			timer->it_signal = sig;
 			hlist_add_head_rcu(&timer->t_hash, head);
 			spin_unlock(&hash_lock);
 			return id;
@@ -406,8 +417,7 @@ static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 
 	/*
 	 * Add the timer to the hash table. The timer is not yet valid
-	 * because new_timer::it_signal is still NULL. The timer id is also
-	 * not yet visible to user space.
+	 * after insertion, but has a unique ID allocated.
 	 */
 	new_timer_id = posix_timer_add(new_timer);
 	if (new_timer_id < 0) {
@@ -415,7 +425,6 @@ static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 		return new_timer_id;
 	}
 
-	new_timer->it_id = (timer_t) new_timer_id;
 	new_timer->it_clock = which_clock;
 	new_timer->kclock = kc;
 	new_timer->it_overrun = -1LL;
@@ -453,7 +462,7 @@ static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 	}
 	/*
 	 * After succesful copy out, the timer ID is visible to user space
-	 * now but not yet valid because new_timer::signal is still NULL.
+	 * now but not yet valid until it's hashed to the process list.
 	 *
 	 * Complete the initialization with the clock specific create
 	 * callback.
@@ -462,11 +471,15 @@ static int do_timer_create(clockid_t which_clock, struct sigevent *event,
 	if (error)
 		goto out;
 
-	spin_lock_irq(&current->sighand->siglock);
-	/* This makes the timer valid in the hash table */
-	WRITE_ONCE(new_timer->it_signal, current->signal);
+	spin_lock_irq(&new_timer->it_lock);
+	spin_lock(&current->sighand->siglock);
 	hlist_add_head(&new_timer->list, &current->signal->posix_timers);
-	spin_unlock_irq(&current->sighand->siglock);
+	spin_unlock(&current->sighand->siglock);
+	/*
+	 * Release initialization and ->timer_create() job to subsequent
+	 * lock_timer()
+	 */
+	spin_unlock_irq(&new_timer->it_lock);
 	/*
 	 * After unlocking sighand::siglock @new_timer is subject to
 	 * concurrent removal and cannot be touched anymore
@@ -526,7 +539,7 @@ static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 	 * rest of the initialization succeeded.
 	 *
 	 * Timer destruction happens in steps:
-	 *  1) Set timr::it_signal to NULL with timr::it_lock held
+	 *  1) Delete timr::list under timr::it_lock held
 	 *  2) Release timr::it_lock
 	 *  3) Remove from the hash under hash_lock
 	 *  4) Put the reference count.
@@ -551,10 +564,12 @@ static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 	if (timr) {
 		spin_lock_irqsave(&timr->it_lock, *flags);
 		/*
-		 * Validate under timr::it_lock that timr::it_signal is
-		 * still valid. Pairs with #1 above.
+		 * Validate under timr::it_lock that the timer hasn't been
+		 * deleted. If so, timr::it_signal isn't expected to change
+		 * within the same RCU critical section. Pairs with #1 above.
 		 */
-		if (timr->it_signal == current->signal) {
+		if (hlist_unhashed(&timr->list) ||
+		    WARN_ON_ONCE(timr->it_signal != current->signal)) {
 			rcu_read_unlock();
 			return timr;
 		}
@@ -1009,7 +1024,7 @@ SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 	}
 
 	spin_lock(&current->sighand->siglock);
-	hlist_del(&timer->list);
+	hlist_del_init(&timer->list);
 	posix_timer_cleanup_ignored(timer);
 	/*
 	 * A concurrent lookup could check timer::it_signal lockless. It

Re: [patch V2 01/17] posix-timers: Initialise timer before adding it to the hash table

[Thomas Gleixner]

On Wed, Mar 05 2025 at 18:25, Frederic Weisbecker wrote:
> Le Sun, Mar 02, 2025 at 08:36:44PM +0100, Thomas Gleixner a écrit :
> Looking at this more or less lockless whole thing again, is the
> ordering between creation and subsequent operations sufficiently guaranteed?
>
>     T0                                                T1
> ---------                                             -----------
> do_timer_create()
>     posix_timer_add()
>         spin_lock(hash_lock)
>         // A
>         timer->it_id = ...
>         spin_unlock(hash_lock)
>     // Initialize timer fields
>     // B
>     new_timer->.... = ....
>     common_timer_create()
>         // C
>         hrtimer_init()
>     spin_lock(current->sighand)
>     // D
>     WRITE_ONCE(new_timer->it_signal, current->signal)
>     spin_unlock(current->sighand)
>                                                       do_timer_settime()
>                                                           lock_timer()
>                                                               // observes A && D
>                                                               posix_timer_by_id()
>                                                               spin_lock_irqsave(&timr->it_lock)
>                                                               // recheck ok
>                                                               if (timr->it_signal == current->signal)
>                                                                   return timr
>                                                               common_timer_get()
>                                                                   // fiddle with timer fields
>                                                                   // but doesn't observe B
>                                                                   // for example doesn't observe SIGEV_NONE
>                                                                   sig_none = timr->it_sigev_notify == SIGEV_NONE;
>                                                                   ...
>                                                                   // doesn't observe C
>                                                                   // hrtimer_init() isn't visible yet
>                                                                   // It might mess up after the hrtimer_start()
>                                                                   hrtimer_start()

Pretty far fetched and I did not think it fully through whether it can
really happen. But that's trivial enough to solve without this
hlist_hashed() indirection:

+      spin_lock(new_timer->lock);
       spin_lock(current->sighand);
       WRITE_ONCE(new_timer->it_signal, current->signal);
       spin_unlock(current->sighand);
+      spin_unlock(new_timer->lock);

Simply because the release of timer::lock guarantees that the memory
operations before the release have been completed before the release
completes.

Consequently the other CPU must observe a consistent set A - D after it
acquired the lock.

No?

Thanks,

        tglx

Re: [patch V2 01/17] posix-timers: Initialise timer before adding it to the hash table

[Frederic Weisbecker]

On Thu, Mar 06, 2025 at 09:10:09AM +0100, Thomas Gleixner wrote:
> On Wed, Mar 05 2025 at 18:25, Frederic Weisbecker wrote:
> > Le Sun, Mar 02, 2025 at 08:36:44PM +0100, Thomas Gleixner a écrit :
> > Looking at this more or less lockless whole thing again, is the
> > ordering between creation and subsequent operations sufficiently guaranteed?
> >
> >     T0                                                T1
> > ---------                                             -----------
> > do_timer_create()
> >     posix_timer_add()
> >         spin_lock(hash_lock)
> >         // A
> >         timer->it_id = ...
> >         spin_unlock(hash_lock)
> >     // Initialize timer fields
> >     // B
> >     new_timer->.... = ....
> >     common_timer_create()
> >         // C
> >         hrtimer_init()
> >     spin_lock(current->sighand)
> >     // D
> >     WRITE_ONCE(new_timer->it_signal, current->signal)
> >     spin_unlock(current->sighand)
> >                                                       do_timer_settime()
> >                                                           lock_timer()
> >                                                               // observes A && D
> >                                                               posix_timer_by_id()
> >                                                               spin_lock_irqsave(&timr->it_lock)
> >                                                               // recheck ok
> >                                                               if (timr->it_signal == current->signal)
> >                                                                   return timr
> >                                                               common_timer_get()
> >                                                                   // fiddle with timer fields
> >                                                                   // but doesn't observe B
> >                                                                   // for example doesn't observe SIGEV_NONE
> >                                                                   sig_none = timr->it_sigev_notify == SIGEV_NONE;
> >                                                                   ...
> >                                                                   // doesn't observe C
> >                                                                   // hrtimer_init() isn't visible yet
> >                                                                   // It might mess up after the hrtimer_start()
> >                                                                   hrtimer_start()
> 
> Pretty far fetched and I did not think it fully through whether it can
> really happen. But that's trivial enough to solve without this
> hlist_hashed() indirection:
> 
> +      spin_lock(new_timer->lock);
>        spin_lock(current->sighand);
>        WRITE_ONCE(new_timer->it_signal, current->signal);
>        spin_unlock(current->sighand);
> +      spin_unlock(new_timer->lock);
> 
> Simply because the release of timer::lock guarantees that the memory
> operations before the release have been completed before the release
> completes.
> 
> Consequently the other CPU must observe a consistent set A - D after it
> acquired the lock.
> 
> No?

Right that's why I added that in my proposed fix and it's the really important
part.

As for the rest, it's just a matter of taste. Checking
timer::it_signal && timer::list instead of flipping a bit in a pointer and
rely on magic comparison seemed to me clearer and easier to follow but that's
just a proposal, I won't mind either way.

Thanks.

> 
> Thanks,
> 
>         tglx

Re: [patch V2 01/17] posix-timers: Initialise timer before adding it to the hash table

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:44PM +0100, Thomas Gleixner a écrit :
> From: Eric Dumazet <edumazet@google.com>
> 
> A timer is only valid in the hashtable when both timer::it_signal and
> timer::it_id are set to their final values, but timers are added without
> those values being set.
> 
> The timer ID is allocated when the timer is added to the hash in invalid
> state. The ID is taken from a monotonically increasing per process counter
> which wraps around after reaching INT_MAX. The hash insertion validates
> that there is no timer with the allocated ID in the hash table which
> belongs to the same process. That opens a mostly theoretical race condition:
> 
> If other threads of the same process manage to create/delete timers in
> rapid succession before the newly created timer is fully initialized and
> wrap around to the timer ID which was handed out, then a duplicate timer ID
> will be inserted into the hash table.
> 
> Prevent this by:
> 
>   1) Setting timer::it_id before inserting the timer into the hashtable.
>  
>   2) Storing the signal pointer in timer::it_signal with bit 0 set before
>      inserting it into the hashtable.
> 
>      Bit 0 acts as a invalid bit, which means that the regular lookup for
>      sys_timer_*() will fail the comparison with the signal pointer.
> 
>      But the lookup on insertion masks out bit 0 and can therefore detect a
>      timer which is not yet valid, but allocated in the hash table.  Bit 0
>      in the pointer is cleared once the initialization of the timer
>      completed.
> 
> [ tglx: Fold ID and signal iniitializaion into one patch and massage change
>   	log and comments. ]
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Link: https://lore.kernel.org/all/20250219125522.2535263-3-edumazet@google.com

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

[patch V2 02/17] posix-timers: Add cond_resched() to posix_timer_add() search loop

[Thomas Gleixner]

From: Eric Dumazet <edumazet@google.com>

With a large number of POSIX timers the search for a valid ID might cause a
soft lockup on PREEMPT_NONE/VOLUNTARY kernels.

Add cond_resched() to the loop to prevent that.

[ tglx: Split out from Eric's series ]

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250214135911.2037402-2-edumazet@google.com
---
 kernel/time/posix-timers.c |    1 +
 1 file changed, 1 insertion(+)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -144,6 +144,7 @@ static int posix_timer_add(struct k_itim
 			return id;
 		}
 		spin_unlock(&hash_lock);
+		cond_resched();
 	}
 	/* POSIX return code when no timer ID could be allocated */
 	return -EAGAIN;

Re: [patch V2 02/17] posix-timers: Add cond_resched() to posix_timer_add() search loop

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:46PM +0100, Thomas Gleixner a écrit :
> From: Eric Dumazet <edumazet@google.com>
> 
> With a large number of POSIX timers the search for a valid ID might cause a
> soft lockup on PREEMPT_NONE/VOLUNTARY kernels.
> 
> Add cond_resched() to the loop to prevent that.
> 
> [ tglx: Split out from Eric's series ]
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Link: https://lore.kernel.org/all/20250214135911.2037402-2-edumazet@google.com

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

[patch V2 03/17] posix-timers: Cleanup includes

[Thomas Gleixner]

Remove pointless includes and sort the remaining ones alphabetically.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/time/posix-timers.c |   26 ++++++++++----------------
 1 file changed, 10 insertions(+), 16 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -9,28 +9,22 @@
  *
  * These are all the functions necessary to implement POSIX clocks & timers
  */
-#include <linux/mm.h>
-#include <linux/interrupt.h>
-#include <linux/slab.h>
-#include <linux/time.h>
-#include <linux/mutex.h>
-#include <linux/sched/task.h>
-
-#include <linux/uaccess.h>
-#include <linux/list.h>
-#include <linux/init.h>
+#include <linux/compat.h>
 #include <linux/compiler.h>
 #include <linux/hash.h>
+#include <linux/hashtable.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/nospec.h>
 #include <linux/posix-clock.h>
 #include <linux/posix-timers.h>
+#include <linux/sched/task.h>
+#include <linux/slab.h>
 #include <linux/syscalls.h>
-#include <linux/wait.h>
-#include <linux/workqueue.h>
-#include <linux/export.h>
-#include <linux/hashtable.h>
-#include <linux/compat.h>
-#include <linux/nospec.h>
+#include <linux/time.h>
 #include <linux/time_namespace.h>
+#include <linux/uaccess.h>
 
 #include "timekeeping.h"
 #include "posix-timers.h"

Re: [patch V2 03/17] posix-timers: Cleanup includes

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:47PM +0100, Thomas Gleixner a écrit :
> Remove pointless includes and sort the remaining ones alphabetically.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Acked-by: Frederic Weisbecker <frederic@kernel.org>

[patch V2 04/17] posix-timers: Remove a few paranoid warnings

[Thomas Gleixner]

Warnings about a non-initialized timer or non-existing callbacks are just
useful for implementing new posix clocks, but there a NULL pointer
dereference is expected anyway. :)

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: New patch
---
 kernel/time/posix-timers.c |   37 ++++++++-----------------------------
 1 file changed, 8 insertions(+), 29 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -675,7 +675,6 @@ void common_timer_get(struct k_itimer *t
 
 static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 {
-	const struct k_clock *kc;
 	struct k_itimer *timr;
 	unsigned long flags;
 	int ret = 0;
@@ -685,11 +684,7 @@ static int do_timer_gettime(timer_t time
 		return -EINVAL;
 
 	memset(setting, 0, sizeof(*setting));
-	kc = timr->kclock;
-	if (WARN_ON_ONCE(!kc || !kc->timer_get))
-		ret = -EINVAL;
-	else
-		kc->timer_get(timr, setting);
+	timr->kclock->timer_get(timr, setting);
 
 	unlock_timer(timr, flags);
 	return ret;
@@ -817,7 +812,6 @@ static void common_timer_wait_running(st
 static struct k_itimer *timer_wait_running(struct k_itimer *timer,
 					   unsigned long *flags)
 {
-	const struct k_clock *kc = READ_ONCE(timer->kclock);
 	timer_t timer_id = READ_ONCE(timer->it_id);
 
 	/* Prevent kfree(timer) after dropping the lock */
@@ -828,8 +822,7 @@ static struct k_itimer *timer_wait_runni
 	 * kc->timer_wait_running() might drop RCU lock. So @timer
 	 * cannot be touched anymore after the function returns!
 	 */
-	if (!WARN_ON_ONCE(!kc->timer_wait_running))
-		kc->timer_wait_running(timer);
+	timer->kclock->timer_wait_running(timer);
 
 	rcu_read_unlock();
 	/* Relock the timer. It might be not longer hashed. */
@@ -892,7 +885,6 @@ static int do_timer_settime(timer_t time
 			    struct itimerspec64 *new_spec64,
 			    struct itimerspec64 *old_spec64)
 {
-	const struct k_clock *kc;
 	struct k_itimer *timr;
 	unsigned long flags;
 	int error;
@@ -915,11 +907,7 @@ static int do_timer_settime(timer_t time
 	/* Prevent signal delivery and rearming. */
 	timr->it_signal_seq++;
 
-	kc = timr->kclock;
-	if (WARN_ON_ONCE(!kc || !kc->timer_set))
-		error = -EINVAL;
-	else
-		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
+	error = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
 
 	if (error == TIMER_RETRY) {
 		// We already got the old time...
@@ -1001,18 +989,6 @@ static inline void posix_timer_cleanup_i
 	}
 }
 
-static inline int timer_delete_hook(struct k_itimer *timer)
-{
-	const struct k_clock *kc = timer->kclock;
-
-	/* Prevent signal delivery and rearming. */
-	timer->it_signal_seq++;
-
-	if (WARN_ON_ONCE(!kc || !kc->timer_del))
-		return -EINVAL;
-	return kc->timer_del(timer);
-}
-
 /* Delete a POSIX.1b interval timer. */
 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 {
@@ -1025,7 +1001,10 @@ SYSCALL_DEFINE1(timer_delete, timer_t, t
 	if (!timer)
 		return -EINVAL;
 
-	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
+	/* Prevent signal delivery and rearming. */
+	timer->it_signal_seq++;
+
+	if (unlikely(timer->kclock->timer_del(timer) == TIMER_RETRY)) {
 		/* Unlocks and relocks the timer if it still exists */
 		timer = timer_wait_running(timer, &flags);
 		goto retry_delete;
@@ -1071,7 +1050,7 @@ static void itimer_delete(struct k_itime
 	 * mechanism. Worse, that timer mechanism might run the expiry
 	 * function concurrently.
 	 */
-	if (timer_delete_hook(timer) == TIMER_RETRY) {
+	if (timer->kclock->timer_del(timer) == TIMER_RETRY) {
 		/*
 		 * Timer is expired concurrently, prevent livelocks
 		 * and pointless spinning on RT.

Re: [patch V2 04/17] posix-timers: Remove a few paranoid warnings

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:49PM +0100, Thomas Gleixner a écrit :
> Warnings about a non-initialized timer or non-existing callbacks are just
> useful for implementing new posix clocks, but there a NULL pointer
> dereference is expected anyway. :)
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
> V2: New patch
> ---
>  kernel/time/posix-timers.c |   37 ++++++++-----------------------------
>  1 file changed, 8 insertions(+), 29 deletions(-)
> 
> --- a/kernel/time/posix-timers.c
> +++ b/kernel/time/posix-timers.c
> @@ -675,7 +675,6 @@ void common_timer_get(struct k_itimer *t
>  
>  static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
>  {
> -	const struct k_clock *kc;
>  	struct k_itimer *timr;
>  	unsigned long flags;
>  	int ret = 0;
> @@ -685,11 +684,7 @@ static int do_timer_gettime(timer_t time
>  		return -EINVAL;
>  
>  	memset(setting, 0, sizeof(*setting));
> -	kc = timr->kclock;
> -	if (WARN_ON_ONCE(!kc || !kc->timer_get))
> -		ret = -EINVAL;
> -	else
> -		kc->timer_get(timr, setting);
> +	timr->kclock->timer_get(timr, setting);
>  
>  	unlock_timer(timr, flags);
>  	return ret;
> @@ -817,7 +812,6 @@ static void common_timer_wait_running(st
>  static struct k_itimer *timer_wait_running(struct k_itimer *timer,
>  					   unsigned long *flags)
>  {
> -	const struct k_clock *kc = READ_ONCE(timer->kclock);
>  	timer_t timer_id = READ_ONCE(timer->it_id);
>  
>  	/* Prevent kfree(timer) after dropping the lock */
> @@ -828,8 +822,7 @@ static struct k_itimer *timer_wait_runni
>  	 * kc->timer_wait_running() might drop RCU lock. So @timer
>  	 * cannot be touched anymore after the function returns!
>  	 */
> -	if (!WARN_ON_ONCE(!kc->timer_wait_running))
> -		kc->timer_wait_running(timer);
> +	timer->kclock->timer_wait_running(timer);
>  
>  	rcu_read_unlock();
>  	/* Relock the timer. It might be not longer hashed. */
> @@ -892,7 +885,6 @@ static int do_timer_settime(timer_t time
>  			    struct itimerspec64 *new_spec64,
>  			    struct itimerspec64 *old_spec64)
>  {
> -	const struct k_clock *kc;
>  	struct k_itimer *timr;
>  	unsigned long flags;
>  	int error;
> @@ -915,11 +907,7 @@ static int do_timer_settime(timer_t time
>  	/* Prevent signal delivery and rearming. */
>  	timr->it_signal_seq++;
>  
> -	kc = timr->kclock;
> -	if (WARN_ON_ONCE(!kc || !kc->timer_set))
> -		error = -EINVAL;
> -	else
> -		error = kc->timer_set(timr, tmr_flags, new_spec64, old_spec64);
> +	error = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
>  
>  	if (error == TIMER_RETRY) {
>  		// We already got the old time...
> @@ -1001,18 +989,6 @@ static inline void posix_timer_cleanup_i
>  	}
>  }
>  
> -static inline int timer_delete_hook(struct k_itimer *timer)
> -{
> -	const struct k_clock *kc = timer->kclock;
> -
> -	/* Prevent signal delivery and rearming. */
> -	timer->it_signal_seq++;
> -
> -	if (WARN_ON_ONCE(!kc || !kc->timer_del))
> -		return -EINVAL;
> -	return kc->timer_del(timer);
> -}
> -
>  /* Delete a POSIX.1b interval timer. */
>  SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
>  {
> @@ -1025,7 +1001,10 @@ SYSCALL_DEFINE1(timer_delete, timer_t, t
>  	if (!timer)
>  		return -EINVAL;
>  
> -	if (unlikely(timer_delete_hook(timer) == TIMER_RETRY)) {
> +	/* Prevent signal delivery and rearming. */
> +	timer->it_signal_seq++;
> +
> +	if (unlikely(timer->kclock->timer_del(timer) == TIMER_RETRY)) {
>  		/* Unlocks and relocks the timer if it still exists */
>  		timer = timer_wait_running(timer, &flags);
>  		goto retry_delete;
> @@ -1071,7 +1050,7 @@ static void itimer_delete(struct k_itime
>  	 * mechanism. Worse, that timer mechanism might run the expiry
>  	 * function concurrently.
>  	 */
> -	if (timer_delete_hook(timer) == TIMER_RETRY) {
> +	if (timer->kclock->timer_del(timer) == TIMER_RETRY) {

And itimer_delete() doesn't need ->it_signal_seq++ since the task is exiting
and signals are going to be flushed anyway, right...

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>


>  		/*
>  		 * Timer is expired concurrently, prevent livelocks
>  		 * and pointless spinning on RT.
> 

[patch V2 05/17] posix-timers: Remove SLAB_PANIC from kmem cache

[Thomas Gleixner]

There is no need to panic when the posix-timer kmem_cache can't be
created. timer_create() will fail with -ENOMEM and that's it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: New patch
---
 kernel/time/posix-timers.c |   11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -243,9 +243,8 @@ static int posix_get_hrtimer_res(clockid
 
 static __init int init_posix_timers(void)
 {
-	posix_timers_cache = kmem_cache_create("posix_timers_cache",
-					sizeof(struct k_itimer), 0,
-					SLAB_PANIC | SLAB_ACCOUNT, NULL);
+	posix_timers_cache = kmem_cache_create("posix_timers_cache", sizeof(struct k_itimer), 0,
+					       SLAB_ACCOUNT, NULL);
 	return 0;
 }
 __initcall(init_posix_timers);
@@ -371,8 +370,12 @@ static struct pid *good_sigevent(sigeven
 
 static struct k_itimer *alloc_posix_timer(void)
 {
-	struct k_itimer *tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
+	struct k_itimer *tmr;
 
+	if (unlikely(!posix_timers_cache))
+		return NULL;
+
+	tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 	if (!tmr)
 		return tmr;
 

Re: [patch V2 05/17] posix-timers: Remove SLAB_PANIC from kmem cache

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:51PM +0100, Thomas Gleixner a écrit :
> There is no need to panic when the posix-timer kmem_cache can't be
> created. timer_create() will fail with -ENOMEM and that's it.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

[patch V2 06/17] posix-timers: Use guards in a few places

[Thomas Gleixner]

Switch locking and RCU to guards where applicable.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: New patch
---
 kernel/time/posix-timers.c |   86 ++++++++++++++++++++-------------------------
 1 file changed, 39 insertions(+), 47 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -397,9 +397,8 @@ void posixtimer_free_timer(struct k_itim
 
 static void posix_timer_unhash_and_free(struct k_itimer *tmr)
 {
-	spin_lock(&hash_lock);
-	hlist_del_rcu(&tmr->t_hash);
-	spin_unlock(&hash_lock);
+	scoped_guard (spinlock, &hash_lock)
+		hlist_del_rcu(&tmr->t_hash);
 	posixtimer_putref(tmr);
 }
 
@@ -443,9 +442,8 @@ static int do_timer_create(clockid_t whi
 	new_timer->it_overrun = -1LL;
 
 	if (event) {
-		rcu_read_lock();
-		new_timer->it_pid = get_pid(good_sigevent(event));
-		rcu_read_unlock();
+		scoped_guard (rcu)
+			new_timer->it_pid = get_pid(good_sigevent(event));
 		if (!new_timer->it_pid) {
 			error = -EINVAL;
 			goto out;
@@ -484,15 +482,15 @@ static int do_timer_create(clockid_t whi
 	if (error)
 		goto out;
 
-	spin_lock_irq(&current->sighand->siglock);
-	/*
-	 * new_timer::it_signal contains the signal pointer with bit 0 set,
-	 * which makes it invalid for syscall operations. Store the
-	 * unmodified signal pointer to make it valid.
-	 */
-	WRITE_ONCE(new_timer->it_signal, current->signal);
-	hlist_add_head(&new_timer->list, &current->signal->posix_timers);
-	spin_unlock_irq(&current->sighand->siglock);
+	scoped_guard (spinlock_irq, &current->sighand->siglock) {
+		/*
+		 * new_timer::it_signal contains the signal pointer with bit 0 set,
+		 * which makes it invalid for syscall operations. Store the
+		 * unmodified signal pointer to make it valid.
+		 */
+		WRITE_ONCE(new_timer->it_signal, current->signal);
+		hlist_add_head(&new_timer->list, &current->signal->posix_timers);
+	}
 	/*
 	 * After unlocking sighand::siglock @new_timer is subject to
 	 * concurrent removal and cannot be touched anymore
@@ -572,7 +570,7 @@ static struct k_itimer *__lock_timer(tim
 	 * can't change, but timr::it_signal becomes NULL during
 	 * destruction.
 	 */
-	rcu_read_lock();
+	guard(rcu)();
 	timr = posix_timer_by_id(timer_id);
 	if (timr) {
 		spin_lock_irqsave(&timr->it_lock, *flags);
@@ -580,14 +578,10 @@ static struct k_itimer *__lock_timer(tim
 		 * Validate under timr::it_lock that timr::it_signal is
 		 * still valid. Pairs with #1 above.
 		 */
-		if (timr->it_signal == current->signal) {
-			rcu_read_unlock();
+		if (timr->it_signal == current->signal)
 			return timr;
-		}
 		spin_unlock_irqrestore(&timr->it_lock, *flags);
 	}
-	rcu_read_unlock();
-
 	return NULL;
 }
 
@@ -818,16 +812,15 @@ static struct k_itimer *timer_wait_runni
 	timer_t timer_id = READ_ONCE(timer->it_id);
 
 	/* Prevent kfree(timer) after dropping the lock */
-	rcu_read_lock();
-	unlock_timer(timer, *flags);
-
-	/*
-	 * kc->timer_wait_running() might drop RCU lock. So @timer
-	 * cannot be touched anymore after the function returns!
-	 */
-	timer->kclock->timer_wait_running(timer);
+	scoped_guard (rcu) {
+		unlock_timer(timer, *flags);
+		/*
+		 * kc->timer_wait_running() might drop RCU lock. So @timer
+		 * cannot be touched anymore after the function returns!
+		 */
+		timer->kclock->timer_wait_running(timer);
+	}
 
-	rcu_read_unlock();
 	/* Relock the timer. It might be not longer hashed. */
 	return lock_timer(timer_id, flags);
 }
@@ -1013,20 +1006,20 @@ SYSCALL_DEFINE1(timer_delete, timer_t, t
 		goto retry_delete;
 	}
 
-	spin_lock(&current->sighand->siglock);
-	hlist_del(&timer->list);
-	posix_timer_cleanup_ignored(timer);
-	/*
-	 * A concurrent lookup could check timer::it_signal lockless. It
-	 * will reevaluate with timer::it_lock held and observe the NULL.
-	 *
-	 * It must be written with siglock held so that the signal code
-	 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
-	 * which prevents it from moving a pending signal of a deleted
-	 * timer to the ignore list.
-	 */
-	WRITE_ONCE(timer->it_signal, NULL);
-	spin_unlock(&current->sighand->siglock);
+	scoped_guard (spinlock, &current->sighand->siglock) {
+		hlist_del(&timer->list);
+		posix_timer_cleanup_ignored(timer);
+		/*
+		 * A concurrent lookup could check timer::it_signal lockless. It
+		 * will reevaluate with timer::it_lock held and observe the NULL.
+		 *
+		 * It must be written with siglock held so that the signal code
+		 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
+		 * which prevents it from moving a pending signal of a deleted
+		 * timer to the ignore list.
+		 */
+		WRITE_ONCE(timer->it_signal, NULL);
+	}
 
 	unlock_timer(timer, flags);
 	posix_timer_unhash_and_free(timer);
@@ -1099,9 +1092,8 @@ void exit_itimers(struct task_struct *ts
 		return;
 
 	/* Protect against concurrent read via /proc/$PID/timers */
-	spin_lock_irq(&tsk->sighand->siglock);
-	hlist_move_list(&tsk->signal->posix_timers, &timers);
-	spin_unlock_irq(&tsk->sighand->siglock);
+	scoped_guard (spinlock_irq, &tsk->sighand->siglock)
+		hlist_move_list(&tsk->signal->posix_timers, &timers);
 
 	/* The timers are not longer accessible via tsk::signal */
 	while (!hlist_empty(&timers)) {

Re: [patch V2 06/17] posix-timers: Use guards in a few places

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:52PM +0100, Thomas Gleixner a écrit :
> Switch locking and RCU to guards where applicable.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

[patch V2 07/17] posix-timers: Simplify lock/unlock_timer()

[Thomas Gleixner]

Since the integration of sigqueue into the timer struct, lock_timer() is
only used in task context. So taking the lock with irqsave() is not longer
required.

Convert it to use spin_[un]lock_irq().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: New patch
---
 kernel/time/posix-timers.c |   70 ++++++++++++++++++---------------------------
 1 file changed, 29 insertions(+), 41 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -53,14 +53,19 @@ static const struct k_clock clock_realti
 #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
 #endif
 
-static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
+static struct k_itimer *__lock_timer(timer_t timer_id);
 
-#define lock_timer(tid, flags)						   \
-({	struct k_itimer *__timr;					   \
-	__cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags));  \
-	__timr;								   \
+#define lock_timer(tid)							\
+({	struct k_itimer *__timr;					\
+	__cond_lock(&__timr->it_lock, __timr = __lock_timer(tid));	\
+	__timr;								\
 })
 
+static inline void unlock_timer(struct k_itimer *timr)
+{
+	spin_unlock_irq(&timr->it_lock);
+}
+
 static int hash(struct signal_struct *sig, unsigned int nr)
 {
 	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
@@ -144,11 +149,6 @@ static int posix_timer_add(struct k_itim
 	return -EAGAIN;
 }
 
-static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
-{
-	spin_unlock_irqrestore(&timr->it_lock, flags);
-}
-
 static int posix_get_realtime_timespec(clockid_t which_clock, struct timespec64 *tp)
 {
 	ktime_get_real_ts64(tp);
@@ -531,7 +531,7 @@ COMPAT_SYSCALL_DEFINE3(timer_create, clo
 }
 #endif
 
-static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
+static struct k_itimer *__lock_timer(timer_t timer_id)
 {
 	struct k_itimer *timr;
 
@@ -573,14 +573,14 @@ static struct k_itimer *__lock_timer(tim
 	guard(rcu)();
 	timr = posix_timer_by_id(timer_id);
 	if (timr) {
-		spin_lock_irqsave(&timr->it_lock, *flags);
+		spin_lock_irq(&timr->it_lock);
 		/*
 		 * Validate under timr::it_lock that timr::it_signal is
 		 * still valid. Pairs with #1 above.
 		 */
 		if (timr->it_signal == current->signal)
 			return timr;
-		spin_unlock_irqrestore(&timr->it_lock, *flags);
+		spin_unlock_irq(&timr->it_lock);
 	}
 	return NULL;
 }
@@ -673,17 +673,16 @@ void common_timer_get(struct k_itimer *t
 static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 {
 	struct k_itimer *timr;
-	unsigned long flags;
 	int ret = 0;
 
-	timr = lock_timer(timer_id, &flags);
+	timr = lock_timer(timer_id);
 	if (!timr)
 		return -EINVAL;
 
 	memset(setting, 0, sizeof(*setting));
 	timr->kclock->timer_get(timr, setting);
 
-	unlock_timer(timr, flags);
+	unlock_timer(timr);
 	return ret;
 }
 
@@ -739,15 +738,14 @@ SYSCALL_DEFINE2(timer_gettime32, timer_t
 SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 {
 	struct k_itimer *timr;
-	unsigned long flags;
 	int overrun;
 
-	timr = lock_timer(timer_id, &flags);
+	timr = lock_timer(timer_id);
 	if (!timr)
 		return -EINVAL;
 
 	overrun = timer_overrun_to_int(timr);
-	unlock_timer(timr, flags);
+	unlock_timer(timr);
 
 	return overrun;
 }
@@ -806,14 +804,13 @@ static void common_timer_wait_running(st
  * when the task which tries to delete or disarm the timer has preempted
  * the task which runs the expiry in task work context.
  */
-static struct k_itimer *timer_wait_running(struct k_itimer *timer,
-					   unsigned long *flags)
+static struct k_itimer *timer_wait_running(struct k_itimer *timer)
 {
 	timer_t timer_id = READ_ONCE(timer->it_id);
 
 	/* Prevent kfree(timer) after dropping the lock */
 	scoped_guard (rcu) {
-		unlock_timer(timer, *flags);
+		unlock_timer(timer);
 		/*
 		 * kc->timer_wait_running() might drop RCU lock. So @timer
 		 * cannot be touched anymore after the function returns!
@@ -822,7 +819,7 @@ static struct k_itimer *timer_wait_runni
 	}
 
 	/* Relock the timer. It might be not longer hashed. */
-	return lock_timer(timer_id, flags);
+	return lock_timer(timer_id);
 }
 
 /*
@@ -882,7 +879,6 @@ static int do_timer_settime(timer_t time
 			    struct itimerspec64 *old_spec64)
 {
 	struct k_itimer *timr;
-	unsigned long flags;
 	int error;
 
 	if (!timespec64_valid(&new_spec64->it_interval) ||
@@ -892,7 +888,7 @@ static int do_timer_settime(timer_t time
 	if (old_spec64)
 		memset(old_spec64, 0, sizeof(*old_spec64));
 
-	timr = lock_timer(timer_id, &flags);
+	timr = lock_timer(timer_id);
 retry:
 	if (!timr)
 		return -EINVAL;
@@ -909,10 +905,10 @@ static int do_timer_settime(timer_t time
 		// We already got the old time...
 		old_spec64 = NULL;
 		/* Unlocks and relocks the timer if it still exists */
-		timr = timer_wait_running(timr, &flags);
+		timr = timer_wait_running(timr);
 		goto retry;
 	}
-	unlock_timer(timr, flags);
+	unlock_timer(timr);
 
 	return error;
 }
@@ -988,10 +984,7 @@ static inline void posix_timer_cleanup_i
 /* Delete a POSIX.1b interval timer. */
 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 {
-	struct k_itimer *timer;
-	unsigned long flags;
-
-	timer = lock_timer(timer_id, &flags);
+	struct k_itimer *timer = lock_timer(timer_id);
 
 retry_delete:
 	if (!timer)
@@ -1002,7 +995,7 @@ SYSCALL_DEFINE1(timer_delete, timer_t, t
 
 	if (unlikely(timer->kclock->timer_del(timer) == TIMER_RETRY)) {
 		/* Unlocks and relocks the timer if it still exists */
-		timer = timer_wait_running(timer, &flags);
+		timer = timer_wait_running(timer);
 		goto retry_delete;
 	}
 
@@ -1021,7 +1014,7 @@ SYSCALL_DEFINE1(timer_delete, timer_t, t
 		WRITE_ONCE(timer->it_signal, NULL);
 	}
 
-	unlock_timer(timer, flags);
+	unlock_timer(timer);
 	posix_timer_unhash_and_free(timer);
 	return 0;
 }
@@ -1032,12 +1025,7 @@ SYSCALL_DEFINE1(timer_delete, timer_t, t
  */
 static void itimer_delete(struct k_itimer *timer)
 {
-	unsigned long flags;
-
-	/*
-	 * irqsave is required to make timer_wait_running() work.
-	 */
-	spin_lock_irqsave(&timer->it_lock, flags);
+	spin_lock_irq(&timer->it_lock);
 
 retry_delete:
 	/*
@@ -1058,7 +1046,7 @@ static void itimer_delete(struct k_itime
 		 * do_exit() only for the last thread of the thread group.
 		 * So no other task can access and delete that timer.
 		 */
-		if (WARN_ON_ONCE(timer_wait_running(timer, &flags) != timer))
+		if (WARN_ON_ONCE(timer_wait_running(timer) != timer))
 			return;
 
 		goto retry_delete;
@@ -1075,7 +1063,7 @@ static void itimer_delete(struct k_itime
 	 */
 	WRITE_ONCE(timer->it_signal, NULL);
 
-	spin_unlock_irqrestore(&timer->it_lock, flags);
+	spin_unlock_irq(&timer->it_lock);
 	posix_timer_unhash_and_free(timer);
 }
 

Re: [patch V2 07/17] posix-timers: Simplify lock/unlock_timer()

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:54PM +0100, Thomas Gleixner a écrit :
> Since the integration of sigqueue into the timer struct, lock_timer() is
> only used in task context. So taking the lock with irqsave() is not longer
> required.
> 
> Convert it to use spin_[un]lock_irq().
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

[patch V2 08/17] posix-timers: Rework timer removal

[Thomas Gleixner]

sys_timer_delete() and the do_exit() cleanup function itimer_delete() are
doing the same thing, but have needlessly different implementations instead
of sharing the code.

The other oddity of timer deletion is the fact that the timer is not
invalidated before the actual deletion happens, which allows concurrent
lookups to succeed.

That's wrong because a timer which is in the process of being deleted
should not be visible and any actions like signal queueing, delivery and
rearming should not happen once the task, which invoked timer_delete(), has
the timer locked.

Rework the code so that:

   1) The signal queueing and delivery code ignore timers which are marked
      invalid

   2) The deletion implementation between sys_timer_delete() and
      itimer_delete() is shared

   3) The timer is invalidated and removed from the linked lists before
      the deletion callback of the relevant clock is invoked.

      That requires to rework timer_wait_running() as it does a lookup of
      the timer when relocking it at the end. In case of deletion this
      lookup would fail due to the preceding invalidation and the wait loop
      would terminate prematurely.

      But due to the preceding invalidation the timer cannot be accessed by
      other tasks anymore, so there is no way that the timer has been freed
      after the timer lock has been dropped.

      Move the re-validation out of timer_wait_running() and handle it at
      the only other usage site, timer_settime().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Simplify timer_wait_running() locking - PeterZ
---
 include/linux/posix-timers.h |    7 +
 kernel/signal.c              |    2 
 kernel/time/posix-timers.c   |  194 ++++++++++++++++++-------------------------
 3 files changed, 90 insertions(+), 113 deletions(-)

--- a/include/linux/posix-timers.h
+++ b/include/linux/posix-timers.h
@@ -240,6 +240,13 @@ static inline void posixtimer_sigqueue_p
 
 	posixtimer_putref(tmr);
 }
+
+static inline bool posixtimer_valid(const struct k_itimer *timer)
+{
+	unsigned long val = (unsigned long)timer->it_signal;
+
+	return !(val & 0x1UL);
+}
 #else  /* CONFIG_POSIX_TIMERS */
 static inline void posixtimer_sigqueue_getref(struct sigqueue *q) { }
 static inline void posixtimer_sigqueue_putref(struct sigqueue *q) { }
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2092,7 +2092,7 @@ static inline void posixtimer_sig_ignore
 	 * from a non-periodic timer, then just drop the reference
 	 * count. Otherwise queue it on the ignored list.
 	 */
-	if (tmr->it_signal && tmr->it_sig_periodic)
+	if (posixtimer_valid(tmr) && tmr->it_sig_periodic)
 		hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers);
 	else
 		posixtimer_putref(tmr);
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -279,7 +279,7 @@ static bool __posixtimer_deliver_signal(
 	 * since the signal was queued. In either case, don't rearm and
 	 * drop the signal.
 	 */
-	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
+	if (timr->it_signal_seq != timr->it_sigqueue_seq || !posixtimer_valid(timr))
 		return false;
 
 	if (!timr->it_interval || WARN_ON_ONCE(timr->it_status != POSIX_TIMER_REQUEUE_PENDING))
@@ -324,6 +324,9 @@ void posix_timer_queue_signal(struct k_i
 {
 	lockdep_assert_held(&timr->it_lock);
 
+	if (!posixtimer_valid(timr))
+		return;
+
 	timr->it_status = timr->it_interval ? POSIX_TIMER_REQUEUE_PENDING : POSIX_TIMER_DISARMED;
 	posixtimer_send_sigqueue(timr);
 }
@@ -546,11 +549,11 @@ static struct k_itimer *__lock_timer(tim
 	 * The hash lookup and the timers are RCU protected.
 	 *
 	 * Timers are added to the hash in invalid state where
-	 * timr::it_signal == NULL. timer::it_signal is only set after the
-	 * rest of the initialization succeeded.
+	 * timr::it_signal is marked invalid. timer::it_signal is only set
+	 * after the rest of the initialization succeeded.
 	 *
 	 * Timer destruction happens in steps:
-	 *  1) Set timr::it_signal to NULL with timr::it_lock held
+	 *  1) Set timr::it_signal marked invalid with timr::it_lock held
 	 *  2) Release timr::it_lock
 	 *  3) Remove from the hash under hash_lock
 	 *  4) Put the reference count.
@@ -567,8 +570,8 @@ static struct k_itimer *__lock_timer(tim
 	 *
 	 * The lookup validates locklessly that timr::it_signal ==
 	 * current::it_signal and timr::it_id == @timer_id. timr::it_id
-	 * can't change, but timr::it_signal becomes NULL during
-	 * destruction.
+	 * can't change, but timr::it_signal can become invalid during
+	 * destruction, which makes the locked check fail.
 	 */
 	guard(rcu)();
 	timr = posix_timer_by_id(timer_id);
@@ -804,22 +807,13 @@ static void common_timer_wait_running(st
  * when the task which tries to delete or disarm the timer has preempted
  * the task which runs the expiry in task work context.
  */
-static struct k_itimer *timer_wait_running(struct k_itimer *timer)
+static void timer_wait_running(struct k_itimer *timer)
 {
-	timer_t timer_id = READ_ONCE(timer->it_id);
-
-	/* Prevent kfree(timer) after dropping the lock */
-	scoped_guard (rcu) {
-		unlock_timer(timer);
-		/*
-		 * kc->timer_wait_running() might drop RCU lock. So @timer
-		 * cannot be touched anymore after the function returns!
-		 */
-		timer->kclock->timer_wait_running(timer);
-	}
-
-	/* Relock the timer. It might be not longer hashed. */
-	return lock_timer(timer_id);
+	/*
+	 * kc->timer_wait_running() might drop RCU lock. So @timer
+	 * cannot be touched anymore after the function returns!
+	 */
+	timer->kclock->timer_wait_running(timer);
 }
 
 /*
@@ -878,8 +872,7 @@ static int do_timer_settime(timer_t time
 			    struct itimerspec64 *new_spec64,
 			    struct itimerspec64 *old_spec64)
 {
-	struct k_itimer *timr;
-	int error;
+	int ret;
 
 	if (!timespec64_valid(&new_spec64->it_interval) ||
 	    !timespec64_valid(&new_spec64->it_value))
@@ -888,29 +881,36 @@ static int do_timer_settime(timer_t time
 	if (old_spec64)
 		memset(old_spec64, 0, sizeof(*old_spec64));
 
-	timr = lock_timer(timer_id);
-retry:
-	if (!timr)
-		return -EINVAL;
+	for (;;) {
+		struct k_itimer *timr = lock_timer(timer_id);
 
-	if (old_spec64)
-		old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
+		if (!timr)
+			return -EINVAL;
 
-	/* Prevent signal delivery and rearming. */
-	timr->it_signal_seq++;
+		if (old_spec64)
+			old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
 
-	error = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
+		/* Prevent signal delivery and rearming. */
+		timr->it_signal_seq++;
+
+		ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
+		if (ret != TIMER_RETRY) {
+			unlock_timer(timr);
+			break;
+		}
 
-	if (error == TIMER_RETRY) {
-		// We already got the old time...
+		/* Read the old time only once */
 		old_spec64 = NULL;
-		/* Unlocks and relocks the timer if it still exists */
-		timr = timer_wait_running(timr);
-		goto retry;
+		/* Protect the timer from being freed after the lock is dropped */
+		guard(rcu)();
+		unlock_timer(timr);
+		/*
+		 * timer_wait_running() might drop RCU read side protection
+		 * so the timer has to be looked up again!
+		 */
+		timer_wait_running(timr);
 	}
-	unlock_timer(timr);
-
-	return error;
+	return ret;
 }
 
 /* Set a POSIX.1b interval timer */
@@ -981,90 +981,56 @@ static inline void posix_timer_cleanup_i
 	}
 }
 
-/* Delete a POSIX.1b interval timer. */
-SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
+static void posix_timer_delete(struct k_itimer *timer)
 {
-	struct k_itimer *timer = lock_timer(timer_id);
-
-retry_delete:
-	if (!timer)
-		return -EINVAL;
-
-	/* Prevent signal delivery and rearming. */
+	/*
+	 * Invalidate the timer, remove it from the linked list and remove
+	 * it from the ignored list if pending.
+	 *
+	 * The invalidation must be written with siglock held so that the
+	 * signal code observes timer->it_valid == false in do_sigaction(),
+	 * which prevents it from moving a pending signal of a deleted
+	 * timer to the ignore list.
+	 *
+	 * The invalidation also prevents signal queueing, signal delivery
+	 * and therefore rearming from the signal delivery path.
+	 *
+	 * A concurrent lookup can still find the timer in the hash, but it
+	 * will check timer::it_signal with timer::it_lock held and observe
+	 * bit 0 set, which invalidates it. That also prevents the timer ID
+	 * from being handed out before this timer is completely gone.
+	 */
 	timer->it_signal_seq++;
 
-	if (unlikely(timer->kclock->timer_del(timer) == TIMER_RETRY)) {
-		/* Unlocks and relocks the timer if it still exists */
-		timer = timer_wait_running(timer);
-		goto retry_delete;
-	}
-
 	scoped_guard (spinlock, &current->sighand->siglock) {
+		unsigned long sig = (unsigned long)timer->it_signal | 1UL;
+
+		WRITE_ONCE(timer->it_signal, (struct signal_struct *)sig);
 		hlist_del(&timer->list);
 		posix_timer_cleanup_ignored(timer);
-		/*
-		 * A concurrent lookup could check timer::it_signal lockless. It
-		 * will reevaluate with timer::it_lock held and observe the NULL.
-		 *
-		 * It must be written with siglock held so that the signal code
-		 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
-		 * which prevents it from moving a pending signal of a deleted
-		 * timer to the ignore list.
-		 */
-		WRITE_ONCE(timer->it_signal, NULL);
 	}
 
-	unlock_timer(timer);
-	posix_timer_unhash_and_free(timer);
-	return 0;
+	while (timer->kclock->timer_del(timer) == TIMER_RETRY) {
+		guard(rcu)();
+		spin_unlock_irq(&timer->it_lock);
+		timer_wait_running(timer);
+		spin_lock_irq(&timer->it_lock);
+	}
 }
 
-/*
- * Delete a timer if it is armed, remove it from the hash and schedule it
- * for RCU freeing.
- */
-static void itimer_delete(struct k_itimer *timer)
+/* Delete a POSIX.1b interval timer. */
+SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 {
-	spin_lock_irq(&timer->it_lock);
-
-retry_delete:
-	/*
-	 * Even if the timer is not longer accessible from other tasks
-	 * it still might be armed and queued in the underlying timer
-	 * mechanism. Worse, that timer mechanism might run the expiry
-	 * function concurrently.
-	 */
-	if (timer->kclock->timer_del(timer) == TIMER_RETRY) {
-		/*
-		 * Timer is expired concurrently, prevent livelocks
-		 * and pointless spinning on RT.
-		 *
-		 * timer_wait_running() drops timer::it_lock, which opens
-		 * the possibility for another task to delete the timer.
-		 *
-		 * That's not possible here because this is invoked from
-		 * do_exit() only for the last thread of the thread group.
-		 * So no other task can access and delete that timer.
-		 */
-		if (WARN_ON_ONCE(timer_wait_running(timer) != timer))
-			return;
-
-		goto retry_delete;
-	}
-	hlist_del(&timer->list);
-
-	posix_timer_cleanup_ignored(timer);
+	struct k_itimer *timer = lock_timer(timer_id);
 
-	/*
-	 * Setting timer::it_signal to NULL is technically not required
-	 * here as nothing can access the timer anymore legitimately via
-	 * the hash table. Set it to NULL nevertheless so that all deletion
-	 * paths are consistent.
-	 */
-	WRITE_ONCE(timer->it_signal, NULL);
+	if (!timer)
+		return -EINVAL;
 
-	spin_unlock_irq(&timer->it_lock);
+	posix_timer_delete(timer);
+	unlock_timer(timer);
+	/* Remove it from the hash, which frees up the timer ID */
 	posix_timer_unhash_and_free(timer);
+	return 0;
 }
 
 /*
@@ -1075,6 +1041,8 @@ static void itimer_delete(struct k_itime
 void exit_itimers(struct task_struct *tsk)
 {
 	struct hlist_head timers;
+	struct hlist_node *next;
+	struct k_itimer *timer;
 
 	if (hlist_empty(&tsk->signal->posix_timers))
 		return;
@@ -1084,8 +1052,10 @@ void exit_itimers(struct task_struct *ts
 		hlist_move_list(&tsk->signal->posix_timers, &timers);
 
 	/* The timers are not longer accessible via tsk::signal */
-	while (!hlist_empty(&timers)) {
-		itimer_delete(hlist_entry(timers.first, struct k_itimer, list));
+	hlist_for_each_entry_safe(timer, next, &timers, list) {
+		scoped_guard (spinlock_irq, &timer->it_lock)
+			posix_timer_delete(timer);
+		posix_timer_unhash_and_free(timer);
 		cond_resched();
 	}
 

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Pavel Tikhomirov]

   > -/* Delete a POSIX.1b interval timer. */
> -SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
> +static void posix_timer_delete(struct k_itimer *timer)
>   {
> -	struct k_itimer *timer = lock_timer(timer_id);
> -
> -retry_delete:
> -	if (!timer)
> -		return -EINVAL;
> -
> -	/* Prevent signal delivery and rearming. */
> +	/*
> +	 * Invalidate the timer, remove it from the linked list and remove
> +	 * it from the ignored list if pending.
> +	 *
> +	 * The invalidation must be written with siglock held so that the
> +	 * signal code observes timer->it_valid == false in do_sigaction(),
> +	 * which prevents it from moving a pending signal of a deleted
> +	 * timer to the ignore list.
> +	 *
> +	 * The invalidation also prevents signal queueing, signal delivery
> +	 * and therefore rearming from the signal delivery path.
> +	 *
> +	 * A concurrent lookup can still find the timer in the hash, but it
> +	 * will check timer::it_signal with timer::it_lock held and observe
> +	 * bit 0 set, which invalidates it. That also prevents the timer ID
> +	 * from being handed out before this timer is completely gone.
> +	 */
>   	timer->it_signal_seq++;
>   
> -	if (unlikely(timer->kclock->timer_del(timer) == TIMER_RETRY)) {
> -		/* Unlocks and relocks the timer if it still exists */
> -		timer = timer_wait_running(timer);
> -		goto retry_delete;
> -	}
> -
>   	scoped_guard (spinlock, &current->sighand->siglock) {
> +		unsigned long sig = (unsigned long)timer->it_signal | 1UL;
> +
> +		WRITE_ONCE(timer->it_signal, (struct signal_struct *)sig);
>   		hlist_del(&timer->list);
>   		posix_timer_cleanup_ignored(timer);
> -		/*
> -		 * A concurrent lookup could check timer::it_signal lockless. It
> -		 * will reevaluate with timer::it_lock held and observe the NULL.
> -		 *
> -		 * It must be written with siglock held so that the signal code
> -		 * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
> -		 * which prevents it from moving a pending signal of a deleted
> -		 * timer to the ignore list.
> -		 */
> -		WRITE_ONCE(timer->it_signal, NULL);
>   	}
>   
> -	unlock_timer(timer);
> -	posix_timer_unhash_and_free(timer);
> -	return 0;
> +	while (timer->kclock->timer_del(timer) == TIMER_RETRY) {
> +		guard(rcu)();
> +		spin_unlock_irq(&timer->it_lock);

Maybe "guard(spinlock_irq)(&timer->it_lock);", since we already use 
guard constructions everywhere else?

> +		timer_wait_running(timer);
> +		spin_lock_irq(&timer->it_lock);
> +	}
>   }
>

-- 
Best regards, Pavel Tikhomirov
Senior Software Developer, Virtuozzo.

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Pavel Tikhomirov]

On 3/4/25 18:10, Pavel Tikhomirov wrote:
>    > -/* Delete a POSIX.1b interval timer. */
>> -SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
>> +static void posix_timer_delete(struct k_itimer *timer)
>>   {
>> -    struct k_itimer *timer = lock_timer(timer_id);
>> -
>> -retry_delete:
>> -    if (!timer)
>> -        return -EINVAL;
>> -
>> -    /* Prevent signal delivery and rearming. */
>> +    /*
>> +     * Invalidate the timer, remove it from the linked list and remove
>> +     * it from the ignored list if pending.
>> +     *
>> +     * The invalidation must be written with siglock held so that the
>> +     * signal code observes timer->it_valid == false in do_sigaction(),
>> +     * which prevents it from moving a pending signal of a deleted
>> +     * timer to the ignore list.
>> +     *
>> +     * The invalidation also prevents signal queueing, signal delivery
>> +     * and therefore rearming from the signal delivery path.
>> +     *
>> +     * A concurrent lookup can still find the timer in the hash, but it
>> +     * will check timer::it_signal with timer::it_lock held and observe
>> +     * bit 0 set, which invalidates it. That also prevents the timer ID
>> +     * from being handed out before this timer is completely gone.
>> +     */
>>       timer->it_signal_seq++;
>> -    if (unlikely(timer->kclock->timer_del(timer) == TIMER_RETRY)) {
>> -        /* Unlocks and relocks the timer if it still exists */
>> -        timer = timer_wait_running(timer);
>> -        goto retry_delete;
>> -    }
>> -
>>       scoped_guard (spinlock, &current->sighand->siglock) {
>> +        unsigned long sig = (unsigned long)timer->it_signal | 1UL;
>> +
>> +        WRITE_ONCE(timer->it_signal, (struct signal_struct *)sig);
>>           hlist_del(&timer->list);
>>           posix_timer_cleanup_ignored(timer);
>> -        /*
>> -         * A concurrent lookup could check timer::it_signal lockless. It
>> -         * will reevaluate with timer::it_lock held and observe the 
>> NULL.
>> -         *
>> -         * It must be written with siglock held so that the signal code
>> -         * observes timer->it_signal == NULL in do_sigaction(SIG_IGN),
>> -         * which prevents it from moving a pending signal of a deleted
>> -         * timer to the ignore list.
>> -         */
>> -        WRITE_ONCE(timer->it_signal, NULL);
>>       }
>> -    unlock_timer(timer);
>> -    posix_timer_unhash_and_free(timer);
>> -    return 0;
>> +    while (timer->kclock->timer_del(timer) == TIMER_RETRY) {
>> +        guard(rcu)();
>> +        spin_unlock_irq(&timer->it_lock);
> 
> Maybe "guard(spinlock_irq)(&timer->it_lock);", since we already use 
> guard constructions everywhere else?

Sorry I'm just stupid. Please ignore.

> 
>> +        timer_wait_running(timer);
>> +        spin_lock_irq(&timer->it_lock);
>> +    }
>>   }
>>
> 

-- 
Best regards, Pavel Tikhomirov
Senior Software Developer, Virtuozzo.

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Thomas Gleixner]

On Tue, Mar 04 2025 at 18:20, Pavel Tikhomirov wrote:
> On 3/4/25 18:10, Pavel Tikhomirov wrote:
>>> +    while (timer->kclock->timer_del(timer) == TIMER_RETRY) {
>>> +        guard(rcu)();
>>> +        spin_unlock_irq(&timer->it_lock);
>> 
>> Maybe "guard(spinlock_irq)(&timer->it_lock);", since we already use 
>> guard constructions everywhere else?
>
> Sorry I'm just stupid. Please ignore.

You're not the first one to trip over this :)

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Frederic Weisbecker]

Le Sun, Mar 02, 2025 at 08:36:56PM +0100, Thomas Gleixner a écrit :
> --- a/include/linux/posix-timers.h
> +++ b/include/linux/posix-timers.h
> @@ -240,6 +240,13 @@ static inline void posixtimer_sigqueue_p
>  
>  	posixtimer_putref(tmr);
>  }
> +
> +static inline bool posixtimer_valid(const struct k_itimer *timer)
> +{
> +	unsigned long val = (unsigned long)timer->it_signal;
> +
> +	return !(val & 0x1UL);
> +}
>  #else  /* CONFIG_POSIX_TIMERS */
>  static inline void posixtimer_sigqueue_getref(struct sigqueue *q) { }
>  static inline void posixtimer_sigqueue_putref(struct sigqueue *q) { }
> --- a/kernel/signal.c
> +++ b/kernel/signal.c
> @@ -2092,7 +2092,7 @@ static inline void posixtimer_sig_ignore
>  	 * from a non-periodic timer, then just drop the reference
>  	 * count. Otherwise queue it on the ignored list.
>  	 */
> -	if (tmr->it_signal && tmr->it_sig_periodic)
> +	if (posixtimer_valid(tmr) && tmr->it_sig_periodic)
>  		hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers);
>  	else
>  		posixtimer_putref(tmr);
> --- a/kernel/time/posix-timers.c
> +++ b/kernel/time/posix-timers.c
> @@ -279,7 +279,7 @@ static bool __posixtimer_deliver_signal(
>  	 * since the signal was queued. In either case, don't rearm and
>  	 * drop the signal.
>  	 */
> -	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
> +	if (timr->it_signal_seq != timr->it_sigqueue_seq || !posixtimer_valid(timr))

Hmm, should it still warn here? ie: WARN_ON_ONCE(!posixtimer_valid(timr)) ?

Thanks.

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Thomas Gleixner]

On Sat, Mar 08 2025 at 00:03, Frederic Weisbecker wrote:
> Le Sun, Mar 02, 2025 at 08:36:56PM +0100, Thomas Gleixner a écrit :
>> --- a/include/linux/posix-timers.h
>> +++ b/include/linux/posix-timers.h
>> @@ -240,6 +240,13 @@ static inline void posixtimer_sigqueue_p
>>  
>>  	posixtimer_putref(tmr);
>>  }
>> +
>> +static inline bool posixtimer_valid(const struct k_itimer *timer)
>> +{
>> +	unsigned long val = (unsigned long)timer->it_signal;
>> +
>> +	return !(val & 0x1UL);
>> +}
>>  #else  /* CONFIG_POSIX_TIMERS */
>>  static inline void posixtimer_sigqueue_getref(struct sigqueue *q) { }
>>  static inline void posixtimer_sigqueue_putref(struct sigqueue *q) { }
>> --- a/kernel/signal.c
>> +++ b/kernel/signal.c
>> @@ -2092,7 +2092,7 @@ static inline void posixtimer_sig_ignore
>>  	 * from a non-periodic timer, then just drop the reference
>>  	 * count. Otherwise queue it on the ignored list.
>>  	 */
>> -	if (tmr->it_signal && tmr->it_sig_periodic)
>> +	if (posixtimer_valid(tmr) && tmr->it_sig_periodic)
>>  		hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers);
>>  	else
>>  		posixtimer_putref(tmr);
>> --- a/kernel/time/posix-timers.c
>> +++ b/kernel/time/posix-timers.c
>> @@ -279,7 +279,7 @@ static bool __posixtimer_deliver_signal(
>>  	 * since the signal was queued. In either case, don't rearm and
>>  	 * drop the signal.
>>  	 */
>> -	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
>> +	if (timr->it_signal_seq != timr->it_sigqueue_seq || !posixtimer_valid(timr))
>
> Hmm, should it still warn here? ie: WARN_ON_ONCE(!posixtimer_valid(timr)) ?

No, because the timer is invalidated early now.

Thanks,

        tglx

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Frederic Weisbecker]

Le Sat, Mar 08, 2025 at 09:34:06AM +0100, Thomas Gleixner a écrit :
> On Sat, Mar 08 2025 at 00:03, Frederic Weisbecker wrote:
> > Le Sun, Mar 02, 2025 at 08:36:56PM +0100, Thomas Gleixner a écrit :
> >> --- a/include/linux/posix-timers.h
> >> +++ b/include/linux/posix-timers.h
> >> @@ -240,6 +240,13 @@ static inline void posixtimer_sigqueue_p
> >>  
> >>  	posixtimer_putref(tmr);
> >>  }
> >> +
> >> +static inline bool posixtimer_valid(const struct k_itimer *timer)
> >> +{
> >> +	unsigned long val = (unsigned long)timer->it_signal;
> >> +
> >> +	return !(val & 0x1UL);
> >> +}
> >>  #else  /* CONFIG_POSIX_TIMERS */
> >>  static inline void posixtimer_sigqueue_getref(struct sigqueue *q) { }
> >>  static inline void posixtimer_sigqueue_putref(struct sigqueue *q) { }
> >> --- a/kernel/signal.c
> >> +++ b/kernel/signal.c
> >> @@ -2092,7 +2092,7 @@ static inline void posixtimer_sig_ignore
> >>  	 * from a non-periodic timer, then just drop the reference
> >>  	 * count. Otherwise queue it on the ignored list.
> >>  	 */
> >> -	if (tmr->it_signal && tmr->it_sig_periodic)
> >> +	if (posixtimer_valid(tmr) && tmr->it_sig_periodic)
> >>  		hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers);
> >>  	else
> >>  		posixtimer_putref(tmr);
> >> --- a/kernel/time/posix-timers.c
> >> +++ b/kernel/time/posix-timers.c
> >> @@ -279,7 +279,7 @@ static bool __posixtimer_deliver_signal(
> >>  	 * since the signal was queued. In either case, don't rearm and
> >>  	 * drop the signal.
> >>  	 */
> >> -	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
> >> +	if (timr->it_signal_seq != timr->it_sigqueue_seq || !posixtimer_valid(timr))
> >
> > Hmm, should it still warn here? ie: WARN_ON_ONCE(!posixtimer_valid(timr)) ?
> 
> No, because the timer is invalidated early now.

But the signal can only be queued before posix_timer_delete(). So
if the bit 0 of it_signal has been reset, it_signal_seq must have been
incremented along, right?

And if so then timr->it_signal_seq == timr->it_sigqueue_seq must imply
posixtimer_valid(). And if not we should warn. Or am I missing something?

Thanks.


> 
> Thanks,
> 
>         tglx

Re: [patch V2 08/17] posix-timers: Rework timer removal

[Thomas Gleixner]

On Sat, Mar 08 2025 at 23:48, Frederic Weisbecker wrote:
> Le Sat, Mar 08, 2025 at 09:34:06AM +0100, Thomas Gleixner a écrit :
>> >> --- a/kernel/time/posix-timers.c
>> >> +++ b/kernel/time/posix-timers.c
>> >> @@ -279,7 +279,7 @@ static bool __posixtimer_deliver_signal(
>> >>  	 * since the signal was queued. In either case, don't rearm and
>> >>  	 * drop the signal.
>> >>  	 */
>> >> -	if (timr->it_signal_seq != timr->it_sigqueue_seq || WARN_ON_ONCE(!timr->it_signal))
>> >> +	if (timr->it_signal_seq != timr->it_sigqueue_seq || !posixtimer_valid(timr))
>> >
>> > Hmm, should it still warn here? ie: WARN_ON_ONCE(!posixtimer_valid(timr)) ?
>> 
>> No, because the timer is invalidated early now.
>
> But the signal can only be queued before posix_timer_delete(). So
> if the bit 0 of it_signal has been reset, it_signal_seq must have been
> incremented along, right?
>
> And if so then timr->it_signal_seq == timr->it_sigqueue_seq must imply
> posixtimer_valid(). And if not we should warn. Or am I missing something?

You're just missing the fact, that I managed to confuse myself :)

[patch V2 09/17] posix-timers: Make lock_timer() use guard()

[Thomas Gleixner]

From: Peter Zijlstra <peterz@infradead.org>

The lookup and locking of posix timers requires the same repeating pattern
at all usage sites:

   tmr = lock_timer(tiner_id);
   if (!tmr)
   	return -EINVAL;
   ....
   unlock_timer(tmr);

Solve this with a guard implementation, which works in most places out of
the box except for those, which need to unlock the timer inside the guard
scope.

Though the only places where this matters are timer_delete() and
timer_settime(). In both cases the timer pointer needs to be preserved
across the end of the scope, which is solved by storing the pointer in a
variable outside of the scope.

timer_settime() also has to protect the timer with RCU before unlocking,
which obviously can't use guard(rcu) before leaving the guard scope as that
guard is cleaned up before the unlock. Solve this by providing the RCU
protection open coded.

[ tglx: Made it work and added change log ]

Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by-yet: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250224162103.GD11590@noisy.programming.kicks-ass.net
---
V2: New patch
---
 include/linux/cleanup.h    |   22 ++++++----
 kernel/time/posix-timers.c |   91 +++++++++++++++++----------------------------
 2 files changed, 49 insertions(+), 64 deletions(-)

--- a/include/linux/cleanup.h
+++ b/include/linux/cleanup.h
@@ -291,11 +291,21 @@ static inline class_##_name##_t class_##
 #define __DEFINE_CLASS_IS_CONDITIONAL(_name, _is_cond)	\
 static __maybe_unused const bool class_##_name##_is_conditional = _is_cond
 
-#define DEFINE_GUARD(_name, _type, _lock, _unlock) \
+#define __DEFINE_GUARD_LOCK_PTR(_name, _exp) \
+	static inline void * class_##_name##_lock_ptr(class_##_name##_t *_T) \
+	{ return (void *)(__force unsigned long)*(_exp); }
+
+#define DEFINE_CLASS_IS_GUARD(_name) \
 	__DEFINE_CLASS_IS_CONDITIONAL(_name, false); \
+	__DEFINE_GUARD_LOCK_PTR(_name, _T)
+
+#define DEFINE_CLASS_IS_COND_GUARD(_name) \
+	__DEFINE_CLASS_IS_CONDITIONAL(_name, true); \
+	__DEFINE_GUARD_LOCK_PTR(_name, _T)
+
+#define DEFINE_GUARD(_name, _type, _lock, _unlock) \
 	DEFINE_CLASS(_name, _type, if (_T) { _unlock; }, ({ _lock; _T; }), _type _T); \
-	static inline void * class_##_name##_lock_ptr(class_##_name##_t *_T) \
-	{ return (void *)(__force unsigned long)*_T; }
+	DEFINE_CLASS_IS_GUARD(_name)
 
 #define DEFINE_GUARD_COND(_name, _ext, _condlock) \
 	__DEFINE_CLASS_IS_CONDITIONAL(_name##_ext, true); \
@@ -375,11 +385,7 @@ static inline void class_##_name##_destr
 	if (_T->lock) { _unlock; }					\
 }									\
 									\
-static inline void *class_##_name##_lock_ptr(class_##_name##_t *_T)	\
-{									\
-	return (void *)(__force unsigned long)_T->lock;			\
-}
-
+__DEFINE_GUARD_LOCK_PTR(_name, &_T->lock)
 
 #define __DEFINE_LOCK_GUARD_1(_name, _type, _lock)			\
 static inline class_##_name##_t class_##_name##_constructor(_type *l)	\
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -66,6 +66,14 @@ static inline void unlock_timer(struct k
 	spin_unlock_irq(&timr->it_lock);
 }
 
+#define scoped_timer_get_or_fail(_id)					\
+	scoped_cond_guard(lock_timer, return -EINVAL, _id)
+
+#define scoped_timer				(scope)
+
+DEFINE_CLASS(lock_timer, struct k_itimer *, unlock_timer(_T), __lock_timer(id), timer_t id);
+DEFINE_CLASS_IS_COND_GUARD(lock_timer);
+
 static int hash(struct signal_struct *sig, unsigned int nr)
 {
 	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
@@ -675,18 +683,10 @@ void common_timer_get(struct k_itimer *t
 
 static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 {
-	struct k_itimer *timr;
-	int ret = 0;
-
-	timr = lock_timer(timer_id);
-	if (!timr)
-		return -EINVAL;
-
 	memset(setting, 0, sizeof(*setting));
-	timr->kclock->timer_get(timr, setting);
-
-	unlock_timer(timr);
-	return ret;
+	scoped_timer_get_or_fail(timer_id)
+		scoped_timer->kclock->timer_get(scoped_timer, setting);
+	return 0;
 }
 
 /* Get the time remaining on a POSIX.1b interval timer. */
@@ -740,17 +740,8 @@ SYSCALL_DEFINE2(timer_gettime32, timer_t
  */
 SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 {
-	struct k_itimer *timr;
-	int overrun;
-
-	timr = lock_timer(timer_id);
-	if (!timr)
-		return -EINVAL;
-
-	overrun = timer_overrun_to_int(timr);
-	unlock_timer(timr);
-
-	return overrun;
+	scoped_timer_get_or_fail(timer_id)
+		return timer_overrun_to_int(scoped_timer);
 }
 
 static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
@@ -868,12 +859,9 @@ int common_timer_set(struct k_itimer *ti
 	return 0;
 }
 
-static int do_timer_settime(timer_t timer_id, int tmr_flags,
-			    struct itimerspec64 *new_spec64,
+static int do_timer_settime(timer_t timer_id, int tmr_flags, struct itimerspec64 *new_spec64,
 			    struct itimerspec64 *old_spec64)
 {
-	int ret;
-
 	if (!timespec64_valid(&new_spec64->it_interval) ||
 	    !timespec64_valid(&new_spec64->it_value))
 		return -EINVAL;
@@ -881,36 +869,28 @@ static int do_timer_settime(timer_t time
 	if (old_spec64)
 		memset(old_spec64, 0, sizeof(*old_spec64));
 
-	for (;;) {
-		struct k_itimer *timr = lock_timer(timer_id);
+	for (; ; old_spec64 = NULL) {
+		struct k_itimer *timr;
 
-		if (!timr)
-			return -EINVAL;
+		scoped_timer_get_or_fail(timer_id) {
+			timr = scoped_timer;
 
-		if (old_spec64)
-			old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
+			if (old_spec64)
+				old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
 
-		/* Prevent signal delivery and rearming. */
-		timr->it_signal_seq++;
-
-		ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
-		if (ret != TIMER_RETRY) {
-			unlock_timer(timr);
-			break;
-		}
+			/* Prevent signal delivery and rearming. */
+			timr->it_signal_seq++;
 
-		/* Read the old time only once */
-		old_spec64 = NULL;
-		/* Protect the timer from being freed after the lock is dropped */
-		guard(rcu)();
-		unlock_timer(timr);
-		/*
-		 * timer_wait_running() might drop RCU read side protection
-		 * so the timer has to be looked up again!
-		 */
+			int ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
+			if (ret != TIMER_RETRY)
+				return ret;
+
+			/* Protect the timer from being freed when leaving the lock scope */
+			rcu_read_lock();
+		}
 		timer_wait_running(timr);
+		rcu_read_unlock();
 	}
-	return ret;
 }
 
 /* Set a POSIX.1b interval timer */
@@ -1021,13 +1001,12 @@ static void posix_timer_delete(struct k_
 /* Delete a POSIX.1b interval timer. */
 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 {
-	struct k_itimer *timer = lock_timer(timer_id);
-
-	if (!timer)
-		return -EINVAL;
+	struct k_itimer *timer;
 
-	posix_timer_delete(timer);
-	unlock_timer(timer);
+	scoped_timer_get_or_fail(timer_id) {
+		timer = scoped_timer;
+		posix_timer_delete(timer);
+	}
 	/* Remove it from the hash, which frees up the timer ID */
 	posix_timer_unhash_and_free(timer);
 	return 0;

[patch V2a 09/17] posix-timers: Make lock_timer() use guard()

[Thomas Gleixner]

From: Peter Zijlstra <peterz@infradead.org>

The lookup and locking of posix timers requires the same repeating pattern
at all usage sites:

   tmr = lock_timer(tiner_id);
   if (!tmr)
   	return -EINVAL;
   ....
   unlock_timer(tmr);

Solve this with a guard implementation, which works in most places out of
the box except for those, which need to unlock the timer inside the guard
scope.

Though the only places where this matters are timer_delete() and
timer_settime(). In both cases the timer pointer needs to be preserved
across the end of the scope, which is solved by storing the pointer in a
variable outside of the scope.

timer_settime() also has to protect the timer with RCU before unlocking,
which obviously can't use guard(rcu) before leaving the guard scope as that
guard is cleaned up before the unlock. Solve this by providing the RCU
protection open coded.

[ tglx: Made it work and added change log ]

Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by-yet: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250224162103.GD11590@noisy.programming.kicks-ass.net
---
V2a: Make unlock conditional - 0day
V2: New patch
---
 include/linux/cleanup.h    |   22 ++++++----
 kernel/time/posix-timers.c |   94 +++++++++++++++++----------------------------
 2 files changed, 51 insertions(+), 65 deletions(-)

--- a/include/linux/cleanup.h
+++ b/include/linux/cleanup.h
@@ -291,11 +291,21 @@ static inline class_##_name##_t class_##
 #define __DEFINE_CLASS_IS_CONDITIONAL(_name, _is_cond)	\
 static __maybe_unused const bool class_##_name##_is_conditional = _is_cond
 
-#define DEFINE_GUARD(_name, _type, _lock, _unlock) \
+#define __DEFINE_GUARD_LOCK_PTR(_name, _exp) \
+	static inline void * class_##_name##_lock_ptr(class_##_name##_t *_T) \
+	{ return (void *)(__force unsigned long)*(_exp); }
+
+#define DEFINE_CLASS_IS_GUARD(_name) \
 	__DEFINE_CLASS_IS_CONDITIONAL(_name, false); \
+	__DEFINE_GUARD_LOCK_PTR(_name, _T)
+
+#define DEFINE_CLASS_IS_COND_GUARD(_name) \
+	__DEFINE_CLASS_IS_CONDITIONAL(_name, true); \
+	__DEFINE_GUARD_LOCK_PTR(_name, _T)
+
+#define DEFINE_GUARD(_name, _type, _lock, _unlock) \
 	DEFINE_CLASS(_name, _type, if (_T) { _unlock; }, ({ _lock; _T; }), _type _T); \
-	static inline void * class_##_name##_lock_ptr(class_##_name##_t *_T) \
-	{ return (void *)(__force unsigned long)*_T; }
+	DEFINE_CLASS_IS_GUARD(_name)
 
 #define DEFINE_GUARD_COND(_name, _ext, _condlock) \
 	__DEFINE_CLASS_IS_CONDITIONAL(_name##_ext, true); \
@@ -375,11 +385,7 @@ static inline void class_##_name##_destr
 	if (_T->lock) { _unlock; }					\
 }									\
 									\
-static inline void *class_##_name##_lock_ptr(class_##_name##_t *_T)	\
-{									\
-	return (void *)(__force unsigned long)_T->lock;			\
-}
-
+__DEFINE_GUARD_LOCK_PTR(_name, &_T->lock)
 
 #define __DEFINE_LOCK_GUARD_1(_name, _type, _lock)			\
 static inline class_##_name##_t class_##_name##_constructor(_type *l)	\
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -63,9 +63,18 @@ static struct k_itimer *__lock_timer(tim
 
 static inline void unlock_timer(struct k_itimer *timr)
 {
-	spin_unlock_irq(&timr->it_lock);
+	if (likely((timr)))
+		spin_unlock_irq(&timr->it_lock);
 }
 
+#define scoped_timer_get_or_fail(_id)					\
+	scoped_cond_guard(lock_timer, return -EINVAL, _id)
+
+#define scoped_timer				(scope)
+
+DEFINE_CLASS(lock_timer, struct k_itimer *, unlock_timer(_T), __lock_timer(id), timer_t id);
+DEFINE_CLASS_IS_COND_GUARD(lock_timer);
+
 static int hash(struct signal_struct *sig, unsigned int nr)
 {
 	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
@@ -675,18 +684,10 @@ void common_timer_get(struct k_itimer *t
 
 static int do_timer_gettime(timer_t timer_id,  struct itimerspec64 *setting)
 {
-	struct k_itimer *timr;
-	int ret = 0;
-
-	timr = lock_timer(timer_id);
-	if (!timr)
-		return -EINVAL;
-
 	memset(setting, 0, sizeof(*setting));
-	timr->kclock->timer_get(timr, setting);
-
-	unlock_timer(timr);
-	return ret;
+	scoped_timer_get_or_fail(timer_id)
+		scoped_timer->kclock->timer_get(scoped_timer, setting);
+	return 0;
 }
 
 /* Get the time remaining on a POSIX.1b interval timer. */
@@ -740,17 +741,8 @@ SYSCALL_DEFINE2(timer_gettime32, timer_t
  */
 SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 {
-	struct k_itimer *timr;
-	int overrun;
-
-	timr = lock_timer(timer_id);
-	if (!timr)
-		return -EINVAL;
-
-	overrun = timer_overrun_to_int(timr);
-	unlock_timer(timr);
-
-	return overrun;
+	scoped_timer_get_or_fail(timer_id)
+		return timer_overrun_to_int(scoped_timer);
 }
 
 static void common_hrtimer_arm(struct k_itimer *timr, ktime_t expires,
@@ -868,12 +860,9 @@ int common_timer_set(struct k_itimer *ti
 	return 0;
 }
 
-static int do_timer_settime(timer_t timer_id, int tmr_flags,
-			    struct itimerspec64 *new_spec64,
+static int do_timer_settime(timer_t timer_id, int tmr_flags, struct itimerspec64 *new_spec64,
 			    struct itimerspec64 *old_spec64)
 {
-	int ret;
-
 	if (!timespec64_valid(&new_spec64->it_interval) ||
 	    !timespec64_valid(&new_spec64->it_value))
 		return -EINVAL;
@@ -881,36 +870,28 @@ static int do_timer_settime(timer_t time
 	if (old_spec64)
 		memset(old_spec64, 0, sizeof(*old_spec64));
 
-	for (;;) {
-		struct k_itimer *timr = lock_timer(timer_id);
+	for (; ; old_spec64 = NULL) {
+		struct k_itimer *timr;
 
-		if (!timr)
-			return -EINVAL;
+		scoped_timer_get_or_fail(timer_id) {
+			timr = scoped_timer;
 
-		if (old_spec64)
-			old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
+			if (old_spec64)
+				old_spec64->it_interval = ktime_to_timespec64(timr->it_interval);
 
-		/* Prevent signal delivery and rearming. */
-		timr->it_signal_seq++;
-
-		ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
-		if (ret != TIMER_RETRY) {
-			unlock_timer(timr);
-			break;
-		}
+			/* Prevent signal delivery and rearming. */
+			timr->it_signal_seq++;
 
-		/* Read the old time only once */
-		old_spec64 = NULL;
-		/* Protect the timer from being freed after the lock is dropped */
-		guard(rcu)();
-		unlock_timer(timr);
-		/*
-		 * timer_wait_running() might drop RCU read side protection
-		 * so the timer has to be looked up again!
-		 */
+			int ret = timr->kclock->timer_set(timr, tmr_flags, new_spec64, old_spec64);
+			if (ret != TIMER_RETRY)
+				return ret;
+
+			/* Protect the timer from being freed when leaving the lock scope */
+			rcu_read_lock();
+		}
 		timer_wait_running(timr);
+		rcu_read_unlock();
 	}
-	return ret;
 }
 
 /* Set a POSIX.1b interval timer */
@@ -1021,13 +1002,12 @@ static void posix_timer_delete(struct k_
 /* Delete a POSIX.1b interval timer. */
 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 {
-	struct k_itimer *timer = lock_timer(timer_id);
-
-	if (!timer)
-		return -EINVAL;
+	struct k_itimer *timer;
 
-	posix_timer_delete(timer);
-	unlock_timer(timer);
+	scoped_timer_get_or_fail(timer_id) {
+		timer = scoped_timer;
+		posix_timer_delete(timer);
+	}
 	/* Remove it from the hash, which frees up the timer ID */
 	posix_timer_unhash_and_free(timer);
 	return 0;

[patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Thomas Gleixner]

From: Eric Dumazet <edumazet@google.com>

The global hash_lock protecting the posix timer hash table can be heavily
contended especially when there is an extensive linear search for a timer
ID.

Timer IDs are handed out by monotonically increasing next_posix_timer_id
and then validating that there is no timer with the same ID in the hash
table. Both operations happen with the global hash lock held.

To reduce the hash lock contention the hash will be reworked to a scaled
hash with per bucket locks, which requires to handle the ID counter
lockless.

Prepare for this by making next_posix_timer_id an atomic_t, which can be
used lockless with atomic_inc_return().

[ tglx: Adopted from Eric's series, massaged change log and simplified it ]

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250219125522.2535263-2-edumazet@google.com
---
V2: Use atomic_fetch_inc() - PeterZ
---
 include/linux/sched/signal.h |    2 +-
 kernel/time/posix-timers.c   |   14 +++++---------
 2 files changed, 6 insertions(+), 10 deletions(-)

--- a/include/linux/sched/signal.h
+++ b/include/linux/sched/signal.h
@@ -136,7 +136,7 @@ struct signal_struct {
 #ifdef CONFIG_POSIX_TIMERS
 
 	/* POSIX.1b Interval Timers */
-	unsigned int		next_posix_timer_id;
+	atomic_t		next_posix_timer_id;
 	struct hlist_head	posix_timers;
 	struct hlist_head	ignored_posix_timers;
 
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -118,21 +118,17 @@ static bool posix_timer_hashed(struct hl
 static int posix_timer_add(struct k_itimer *timer)
 {
 	struct signal_struct *sig = current->signal;
-	struct hlist_head *head;
-	unsigned int cnt, id;
 
 	/*
 	 * FIXME: Replace this by a per signal struct xarray once there is
 	 * a plan to handle the resulting CRIU regression gracefully.
 	 */
-	for (cnt = 0; cnt <= INT_MAX; cnt++) {
-		spin_lock(&hash_lock);
-		id = sig->next_posix_timer_id;
-
-		/* Write the next ID back. Clamp it to the positive space */
-		sig->next_posix_timer_id = (id + 1) & INT_MAX;
+	for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {
+		/* Get the next timer ID and clamp it to positive space */
+		unsigned int id = atomic_fetch_inc(&sig->next_posix_timer_id) & INT_MAX;
+		struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
 
-		head = &posix_timers_hashtable[hash(sig, id)];
+		spin_lock(&hash_lock);
 		if (!posix_timer_hashed(head, sig, id)) {
 			/*
 			 * Set the timer ID and the signal pointer to make

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Cyrill Gorcunov]

On Sun, Mar 02, 2025 at 08:36:59PM +0100, Thomas Gleixner wrote:
>  static int posix_timer_add(struct k_itimer *timer)
>  {
...
> +	for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {

Hi, Thomas! This moment slightly confusing -- are we allowed to declare
type inside 'for' statement? (c99 allows, just never seen it before inside
the kernel code). Not a big deal though, thanks a huge for fixing this
problem with timers and criu.

	Cyrill

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Thomas Gleixner]

On Mon, Mar 03 2025 at 23:21, Cyrill Gorcunov wrote:
> On Sun, Mar 02, 2025 at 08:36:59PM +0100, Thomas Gleixner wrote:
>>  static int posix_timer_add(struct k_itimer *timer)
>>  {
> ...
>> +	for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {
>
> Hi, Thomas! This moment slightly confusing -- are we allowed to declare
> type inside 'for' statement? (c99 allows, just never seen it before inside
> the kernel code).

We're slowly moving towards scope based variables. It's not widely used
yet, but it's coming along.

> Not a big deal though, thanks a huge for fixing this problem with
> timers and criu.

Welcome. Some quick validation with CRIU would be appreciated.

Thanks,

        tglx

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Cyrill Gorcunov]

On Mon, Mar 03, 2025 at 10:24:28PM +0100, Thomas Gleixner wrote:
> 
> Welcome. Some quick validation with CRIU would be appreciated.

Just tested in criu: works without problem, both modes -- with new
prctl and without it. Note that I only have ran separate posix-timers
test case, probably virtuozzo team might do more deep tesing.
---
root@fc:/home/cyrill/criu# test/zdtm.py run -t zdtm/static/posix_timers
userns is supported
The kernel is tainted: '512'
=== Run 1/1 ================ zdtm/static/posix_timers
====================== Run zdtm/static/posix_timers in h =======================
Start test
./posix_timers --pidfile=posix_timers.pid --outfile=posix_timers.out
Run criu dump
Run criu restore
Send the 15 signal to  56
Wait for zdtm/static/posix_timers(56) to die for 0.100000
Removing dump/zdtm/static/posix_timers/56
====================== Test zdtm/static/posix_timers PASS ======================

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Thomas Gleixner]

On Tue, Mar 04 2025 at 20:56, Cyrill Gorcunov wrote:
> On Mon, Mar 03, 2025 at 10:24:28PM +0100, Thomas Gleixner wrote:
>> 
>> Welcome. Some quick validation with CRIU would be appreciated.
>
> Just tested in criu: works without problem, both modes -- with new
> prctl and without it. Note that I only have ran separate posix-timers
> test case, probably virtuozzo team might do more deep tesing.

Thank you very much!

      tglx

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Cyrill Gorcunov]

On Tue, Mar 04, 2025 at 09:30:04PM +0100, Thomas Gleixner wrote:
> On Tue, Mar 04 2025 at 20:56, Cyrill Gorcunov wrote:
> > On Mon, Mar 03, 2025 at 10:24:28PM +0100, Thomas Gleixner wrote:
> >> 
> >> Welcome. Some quick validation with CRIU would be appreciated.
> >
> > Just tested in criu: works without problem, both modes -- with new
> > prctl and without it. Note that I only have ran separate posix-timers
> > test case, probably virtuozzo team might do more deep tesing.
> 
> Thank you very much!

Thanks for handling this) Also looking into this series I wonder why can't
we instead of mangling ::it_signal zero bit just use ::it_id with negative
value as a sign of not yet fully initialized timer? This would allow to not
read-modify action while traversing bucket hash chain. I mean we could do

static bool posix_timer_add_at(struct k_itimer *timer, struct signal_struct *sig, unsigned int id)
{
	struct timer_hash_bucket *bucket = hash_bucket(sig, id);

	scoped_guard (spinlock, &bucket->lock) {
		if (!posix_timer_hashed(bucket, sig, id)) {
--->			timer->it_id = -(timer_t)id;
			timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
			hlist_add_head_rcu(&timer->t_hash, &bucket->head);
			return true;
		}
	}
	return false;
}

Then hash traverse won't find the timer until the do_timer_create will do

	scoped_guard (spinlock_irq, &current->sighand->siglock) {
		WRITE_ONCE(new_timer->it_id, abs(new_timer->it_id));
		hlist_add_head_rcu(&new_timer->list, &current->signal->posix_timers);
	}

Or I miss something obvious? (Of course when deleting timer we will have to pass
abs it_id for hash traversing).

It looks that in case of many many timers present in the system traversing hash
in read-modify way might be heavy (though I didn't measure of course).

	Cyrill

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Thomas Gleixner]

On Wed, Mar 05 2025 at 01:16, Cyrill Gorcunov wrote:
> Thanks for handling this) Also looking into this series I wonder why can't
> we instead of mangling ::it_signal zero bit just use ::it_id with negative
> value as a sign of not yet fully initialized timer? This would allow to not
> read-modify action while traversing bucket hash chain. I mean we could do
>
> static bool posix_timer_add_at(struct k_itimer *timer, struct signal_struct *sig, unsigned int id)
> {
> 	struct timer_hash_bucket *bucket = hash_bucket(sig, id);
>
> 	scoped_guard (spinlock, &bucket->lock) {
> 		if (!posix_timer_hashed(bucket, sig, id)) {
> --->			timer->it_id = -(timer_t)id;
> 			timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
> 			hlist_add_head_rcu(&timer->t_hash, &bucket->head);
> 			return true;
> 		}
> 	}
> 	return false;
> }
>
> Then hash traverse won't find the timer until the do_timer_create will do
>
> 	scoped_guard (spinlock_irq, &current->sighand->siglock) {
> 		WRITE_ONCE(new_timer->it_id, abs(new_timer->it_id));
> 		hlist_add_head_rcu(&new_timer->list, &current->signal->posix_timers);
> 	}
>
> Or I miss something obvious? (Of course when deleting timer we will have to pass
> abs it_id for hash traversing).
>
> It looks that in case of many many timers present in the system traversing hash
> in read-modify way might be heavy (though I didn't measure of course).

The traversal does not RMW the timer itself, it unmangles the signal
pointer for comparison in posix_timer_hashed(). posix_timer_by_id() does
straight comparisons. So both only read.

Sure, we can mangle timer ID instead of the signal pointer, but the
outcome is pretty much the same. The only difference is in
posix_timer_hashed(), which must detect a taken timer ID independent of
the timers valid state to prevent collisions.

With the signal pointer mangling we have:

   if ((timer->signal & ~1) == sig && timer->id == id)

and with the negative ID value this becomes:

   if (timer->signal == sig && (timer->id == id || timer->id == -id))

which is obviously worse. You'd need to do:

      timer->id = id | (1 << 31);

and then the posix_timer_hashed() check becomes:

   if (timer->signal == sig && (timer->id & ~(1 << 31)) == id)

Granted, the timer ID mangling spares the AND operation on the signal in
case the timer is not owned by the currrent process, but I doubt that
this is even measurable.

Thanks,

        tglx

Re: [patch V2 10/17] posix-timers: Make signal_struct::next_posix_timer_id an atomic_t

[Cyrill Gorcunov]

On Wed, Mar 05, 2025 at 08:31:21AM +0100, Thomas Gleixner wrote:
...
> 
> The traversal does not RMW the timer itself, it unmangles the signal
> pointer for comparison in posix_timer_hashed(). posix_timer_by_id() does
> straight comparisons. So both only read.

No, I mean that we read the value then allocate a temp value with 0 bit
excluded implicitly, so it is not a straight read, but whatever.

> Sure, we can mangle timer ID instead of the signal pointer, but the
> outcome is pretty much the same. The only difference is in
> posix_timer_hashed(), which must detect a taken timer ID independent of
> the timers valid state to prevent collisions.

Bah, I managed to miss that we need to lookup for not yet fully initialized
timers as well, and indeed it makes no much difference which exactly field
to mangle. Thanks a huge for explanations, Thomas!

	Cyrill

[patch V2 11/17] posix-timers: Improve hash table performance

[Thomas Gleixner]

Eric and Ben reported a significant performance bottleneck on the global
hash, which is used to store posix timers for lookup.

Eric tried to do a lockless validation of a new timer ID before trying to
insert the timer, but that does not solve the problem.

For the non-contended case this is a pointless exercise and for the
contended case this extra lookup just creates enough interleaving that all
tasks can make progress.

There are actually two real solutions to the problem:

  1) Provide a per process (signal struct) xarray storage

  2) Implement a smarter hash like the one in the futex code

#1 works perfectly fine for most cases, but the fact that CRIU enforced a
   linear increasing timer ID to restore timers makes this problematic.

   It's easy enough to create a sparse timer ID space, which amounts very
   fast to a large junk of memory consumed for the xarray. 2048 timers with
   a ID offset of 512 consume more than one megabyte of memory for the
   xarray storage.

#2 The main advantage of the futex hash is that it uses per hash bucket
   locks instead of a global hash lock. Aside of that it is scaled
   according to the number of CPUs at boot time.

Experiments with artifical benchmarks have shown that a scaled hash with
per bucket locks comes pretty close to the xarray performance and in some
scenarios it performes better.

Test 1:

     A single process creates 20000 timers and afterwards invokes
     timer_getoverrun(2) on each of them:

            mainline        Eric   newhash   xarray
create         23 ms       23 ms      9 ms     8 ms
getoverrun     14 ms       14 ms      5 ms     4 ms

Test 2:

     A single process creates 50000 timers and afterwards invokes
     timer_getoverrun(2) on each of them:

            mainline        Eric   newhash   xarray
create         98 ms      219 ms     20 ms    18 ms
getoverrun     62 ms       62 ms     10 ms     9 ms

Test 3:

     A single process creates 100000 timers and afterwards invokes
     timer_getoverrun(2) on each of them:

            mainline        Eric   newhash   xarray
create        313 ms      750 ms     48 ms    33 ms
getoverrun    261 ms      260 ms     20 ms    14 ms

Erics changes create quite some overhead in the create() path due to the
double list walk, as the main issue according to perf is the list walk
itself. With 100k timers each hash bucket contains ~200 timers, which in
the worst case need to be all inspected. The same problem applies for
getoverrun() where the lookup has to walk through the hash buckets to find
the timer it is looking for.

The scaled hash obviously reduces hash collisions and lock contention
significantly. This becomes more prominent with concurrency.

Test 4:

     A process creates 63 threads and all threads wait on a barrier before
     each instance creates 20000 timers and afterwards invokes
     timer_getoverrun(2) on each of them. The threads are pinned on
     seperate CPUs to achive maximum concurrency. The numbers are the
     average times per thread:

            mainline        Eric   newhash   xarray
create     180239 ms    38599 ms    579 ms   813 ms
getoverrun   2645 ms     2642 ms     32 ms     7 ms

Test 5:

     A process forks 63 times and all forks wait on a barrier before each
     instance creates 20000 timers and afterwards invokes
     timer_getoverrun(2) on each of them. The processes are pinned on
     seperate CPUs to achive maximum concurrency. The numbers are the
     average times per process:

            mainline        eric   newhash   xarray
create     157253 ms    40008 ms     83 ms    60 ms
getoverrun   2611 ms     2614 ms     40 ms     4 ms

So clearly the reduction of lock contention with Eric's changes makes a
significant difference for the create() loop, but it does not mitigate the
problem of long list walks, which is clearly visible on the getoverrun()
side because that is purely dominated by the lookup itself. Once the timer
is found, the syscall just reads from the timer structure with no other
locks or code paths involved and returns.

The reason for the difference between the thread and the fork case for the
new hash and the xarray is that both suffer from contention on
sighand::siglock and the xarray suffers additionally from contention on the
xarray lock on insertion.

The only case where the reworked hash slighly outperforms the xarray is a
tight loop which creates and deletes timers.

Test 4:

     A process creates 63 threads and all threads wait on a barrier before
     each instance runs a loop which creates and deletes a timer 100000
     times in a row. The threads are pinned on seperate CPUs to achive
     maximum concurrency. The numbers are the average times per thread:

            mainline        Eric   newhash   xarray
loop	    5917  ms	 5897 ms   5473 ms  7846 ms

Test 5:

     A process forks 63 times and all forks wait on a barrier before each
     each instance runs a loop which creates and deletes a timer 100000
     times in a row. The processes are pinned on seperate CPUs to achive
     maximum concurrency. The numbers are the average times per process:

            mainline        Eric   newhash   xarray
loop	     5137 ms	 7828 ms    891 ms   872 ms

In both test there is not much contention on the hash, but the ucount
accounting for the signal and in the thread case the sighand::siglock
contention (plus the xarray locking) contribute dominantly to the overhead.

As the memory consumption of the xarray in the sparse ID case is
significant, the scaled hash with per bucket locks seems to be the better
overall option. While the xarray has faster lookup times for a large number
of timers, the actual syscall usage, which requires the lookup is not an
extreme hotpath. Most applications utilize signal delivery and all syscalls
except timer_getoverrun(2) are all but cheap.

So implement a scaled hash with per bucket locks, which offers the best
tradeoff between performance and memory consumption.

Reported-by: Eric Dumazet <edumazet@google.com>
Reported-by: Benjamin Segall <bsegall@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Replace hash() by hashbucket(), which returns the bucket pointer.
---
 kernel/time/posix-timers.c |   99 ++++++++++++++++++++++++++++++---------------
 1 file changed, 68 insertions(+), 31 deletions(-)
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -12,10 +12,10 @@
 #include <linux/compat.h>
 #include <linux/compiler.h>
 #include <linux/hash.h>
-#include <linux/hashtable.h>
 #include <linux/init.h>
 #include <linux/interrupt.h>
 #include <linux/list.h>
+#include <linux/memblock.h>
 #include <linux/nospec.h>
 #include <linux/posix-clock.h>
 #include <linux/posix-timers.h>
@@ -40,8 +40,18 @@ static struct kmem_cache *posix_timers_c
  * This allows checkpoint/restore to reconstruct the exact timer IDs for
  * a process.
  */
-static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
-static DEFINE_SPINLOCK(hash_lock);
+struct timer_hash_bucket {
+	spinlock_t		lock;
+	struct hlist_head	head;
+};
+
+static struct {
+	struct timer_hash_bucket	*buckets;
+	unsigned long			bits;
+} __timer_data __ro_after_init __aligned(2*sizeof(long));
+
+#define timer_buckets	(__timer_data.buckets)
+#define timer_hashbits	(__timer_data.bits)
 
 static const struct k_clock * const posix_clocks[];
 static const struct k_clock *clockid_to_kclock(const clockid_t id);
@@ -74,18 +84,18 @@ static inline void unlock_timer(struct k
 DEFINE_CLASS(lock_timer, struct k_itimer *, unlock_timer(_T), __lock_timer(id), timer_t id);
 DEFINE_CLASS_IS_COND_GUARD(lock_timer);
 
-static int hash(struct signal_struct *sig, unsigned int nr)
+static struct timer_hash_bucket *hash_bucket(struct signal_struct *sig, unsigned int nr)
 {
-	return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
+	return &timer_buckets[hash_32(hash32_ptr(sig) ^ nr, timer_hashbits)];
 }
 
 static struct k_itimer *posix_timer_by_id(timer_t id)
 {
 	struct signal_struct *sig = current->signal;
-	struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
+	struct timer_hash_bucket *bucket = hash_bucket(sig, id);
 	struct k_itimer *timer;
 
-	hlist_for_each_entry_rcu(timer, head, t_hash) {
+	hlist_for_each_entry_rcu(timer, &bucket->head, t_hash) {
 		/* timer->it_signal can be set concurrently */
 		if ((READ_ONCE(timer->it_signal) == sig) && (timer->it_id == id))
 			return timer;
@@ -104,11 +114,13 @@ static inline struct signal_struct *posi
 	return (struct signal_struct *)(val & ~1UL);
 }
 
-static bool posix_timer_hashed(struct hlist_head *head, struct signal_struct *sig, timer_t id)
+static bool posix_timer_hashed(struct timer_hash_bucket *bucket, struct signal_struct *sig,
+			       timer_t id)
 {
+	struct hlist_head *head = &bucket->head;
 	struct k_itimer *timer;
 
-	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&hash_lock)) {
+	hlist_for_each_entry_rcu(timer, head, t_hash, lockdep_is_held(&bucket->lock)) {
 		if ((posix_sig_owner(timer) == sig) && (timer->it_id == id))
 			return true;
 	}
@@ -119,34 +131,34 @@ static int posix_timer_add(struct k_itim
 {
 	struct signal_struct *sig = current->signal;
 
-	/*
-	 * FIXME: Replace this by a per signal struct xarray once there is
-	 * a plan to handle the resulting CRIU regression gracefully.
-	 */
 	for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {
 		/* Get the next timer ID and clamp it to positive space */
 		unsigned int id = atomic_fetch_inc(&sig->next_posix_timer_id) & INT_MAX;
-		struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
+		struct timer_hash_bucket *bucket = hash_bucket(sig, id);
 
-		spin_lock(&hash_lock);
-		if (!posix_timer_hashed(head, sig, id)) {
+		scoped_guard (spinlock, &bucket->lock) {
 			/*
-			 * Set the timer ID and the signal pointer to make
-			 * it identifiable in the hash table. The signal
-			 * pointer has bit 0 set to indicate that it is not
-			 * yet fully initialized. posix_timer_hashed()
-			 * masks this bit out, but the syscall lookup fails
-			 * to match due to it being set. This guarantees
-			 * that there can't be duplicate timer IDs handed
-			 * out.
+			 * Validate under the lock as this could have raced
+			 * against another thread ending up with the same
+			 * ID, which is highly unlikely, but possible.
 			 */
-			timer->it_id = (timer_t)id;
-			timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
-			hlist_add_head_rcu(&timer->t_hash, head);
-			spin_unlock(&hash_lock);
-			return id;
+			if (!posix_timer_hashed(bucket, sig, id)) {
+				/*
+				 * Set the timer ID and the signal pointer to make
+				 * it identifiable in the hash table. The signal
+				 * pointer has bit 0 set to indicate that it is not
+				 * yet fully initialized. posix_timer_hashed()
+				 * masks this bit out, but the syscall lookup fails
+				 * to match due to it being set. This guarantees
+				 * that there can't be duplicate timer IDs handed
+				 * out.
+				 */
+				timer->it_id = (timer_t)id;
+				timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
+				hlist_add_head_rcu(&timer->t_hash, &bucket->head);
+				return id;
+			}
 		}
-		spin_unlock(&hash_lock);
 		cond_resched();
 	}
 	/* POSIX return code when no timer ID could be allocated */
@@ -404,7 +416,9 @@ void posixtimer_free_timer(struct k_itim
 
 static void posix_timer_unhash_and_free(struct k_itimer *tmr)
 {
-	scoped_guard (spinlock, &hash_lock)
+	struct timer_hash_bucket *bucket = hash_bucket(posix_sig_owner(tmr), tmr->it_id);
+
+	scoped_guard (spinlock, &bucket->lock)
 		hlist_del_rcu(&tmr->t_hash);
 	posixtimer_putref(tmr);
 }
@@ -1477,3 +1491,26 @@ static const struct k_clock *clockid_to_
 
 	return posix_clocks[array_index_nospec(idx, ARRAY_SIZE(posix_clocks))];
 }
+
+static int __init posixtimer_init(void)
+{
+	unsigned long i, size;
+	unsigned int shift;
+
+	if (IS_ENABLED(CONFIG_BASE_SMALL))
+		size = 512;
+	else
+		size = roundup_pow_of_two(512 * num_possible_cpus());
+
+	timer_buckets = alloc_large_system_hash("posixtimers", sizeof(*timer_buckets),
+						size, 0, 0, &shift, NULL, size, size);
+	size = 1UL << shift;
+	timer_hashbits = ilog2(size);
+
+	for (i = 0; i < size; i++) {
+		spin_lock_init(&timer_buckets[i].lock);
+		INIT_HLIST_HEAD(&timer_buckets[i].head);
+	}
+	return 0;
+}
+core_initcall(posixtimer_init);

[patch V2 12/17] posix-timers: Switch to jhash32()

[Thomas Gleixner]

The hash distribution of hash_32() is suboptimal. jhash32() provides a way
better distribution, which evens out the length of the hash bucket lists,
which in turn avoids large outliers in list walk times.

Due to the sparse ID space (thanks CRIU) there is no guarantee that the
timers will be fully evenly distributed over the hash buckets, but the
behaviour is way better than with hash_32() even for randomly sparse ID
spaces.

For a pathological test case with 64 processes creating and accessing
20000 timers each, this results in a runtime reduction of ~10% and a
significantly reduced runtime variation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: New patch
---
 kernel/time/posix-timers.c |   10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -11,8 +11,8 @@
  */
 #include <linux/compat.h>
 #include <linux/compiler.h>
-#include <linux/hash.h>
 #include <linux/init.h>
+#include <linux/jhash.h>
 #include <linux/interrupt.h>
 #include <linux/list.h>
 #include <linux/memblock.h>
@@ -47,11 +47,11 @@ struct timer_hash_bucket {
 
 static struct {
 	struct timer_hash_bucket	*buckets;
-	unsigned long			bits;
+	unsigned long			mask;
 } __timer_data __ro_after_init __aligned(2*sizeof(long));
 
 #define timer_buckets	(__timer_data.buckets)
-#define timer_hashbits	(__timer_data.bits)
+#define timer_hashmask	(__timer_data.mask)
 
 static const struct k_clock * const posix_clocks[];
 static const struct k_clock *clockid_to_kclock(const clockid_t id);
@@ -86,7 +86,7 @@ DEFINE_CLASS_IS_COND_GUARD(lock_timer);
 
 static struct timer_hash_bucket *hash_bucket(struct signal_struct *sig, unsigned int nr)
 {
-	return &timer_buckets[hash_32(hash32_ptr(sig) ^ nr, timer_hashbits)];
+	return &timer_buckets[jhash2((u32 *)&sig, sizeof(sig) / sizeof(u32), nr) & timer_hashmask];
 }
 
 static struct k_itimer *posix_timer_by_id(timer_t id)
@@ -1505,7 +1505,7 @@ static int __init posixtimer_init(void)
 	timer_buckets = alloc_large_system_hash("posixtimers", sizeof(*timer_buckets),
 						size, 0, 0, &shift, NULL, size, size);
 	size = 1UL << shift;
-	timer_hashbits = ilog2(size);
+	timer_hashmask = size - 1;
 
 	for (i = 0; i < size; i++) {
 		spin_lock_init(&timer_buckets[i].lock);

[patch V2 13/17] posix-timers: Avoid false cacheline sharing

[Thomas Gleixner]

struct k_itimer has the hlist_node, which is used for lookup in the hash
bucket, and the timer lock in the same cache line.

That's obviously bad, if one CPU fiddles with a timer and the other is
walking the hash bucket on which that timer is queued.

Avoid this by restructuring struct k_itimer, so that the read mostly (only
modified during setup and teardown) fields are in the first cache line and
the lock and the rest of the fields which get written to are in cacheline
2-N.

Reduces cacheline contention in a test case of 64 processes creating and
accessing 20000 timers each by almost 30% according to perf.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: New patch
---
 include/linux/posix-timers.h |   21 ++++++++++++---------
 kernel/time/posix-timers.c   |    4 ++--
 2 files changed, 14 insertions(+), 11 deletions(-)

--- a/include/linux/posix-timers.h
+++ b/include/linux/posix-timers.h
@@ -177,23 +177,26 @@ static inline void posix_cputimers_init_
  * @rcu:		RCU head for freeing the timer.
  */
 struct k_itimer {
-	struct hlist_node	list;
-	struct hlist_node	ignored_list;
+	/* 1st cacheline contains read-mostly fields */
 	struct hlist_node	t_hash;
-	spinlock_t		it_lock;
-	const struct k_clock	*kclock;
-	clockid_t		it_clock;
+	struct hlist_node	list;
 	timer_t			it_id;
+	clockid_t		it_clock;
+	int			it_sigev_notify;
+	enum pid_type		it_pid_type;
+	struct signal_struct	*it_signal;
+	const struct k_clock	*kclock;
+
+	/* 2nd cacheline and above contain fields which are modified regularly */
+	spinlock_t		it_lock;
 	int			it_status;
 	bool			it_sig_periodic;
 	s64			it_overrun;
 	s64			it_overrun_last;
 	unsigned int		it_signal_seq;
 	unsigned int		it_sigqueue_seq;
-	int			it_sigev_notify;
-	enum pid_type		it_pid_type;
 	ktime_t			it_interval;
-	struct signal_struct	*it_signal;
+	struct hlist_node	ignored_list;
 	union {
 		struct pid		*it_pid;
 		struct task_struct	*it_process;
@@ -210,7 +213,7 @@ struct k_itimer {
 		} alarm;
 	} it;
 	struct rcu_head		rcu;
-};
+} ____cacheline_aligned_in_smp;
 
 void run_posix_cpu_timers(void);
 void posix_cpu_timers_exit(struct task_struct *task);
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -259,8 +259,8 @@ static int posix_get_hrtimer_res(clockid
 
 static __init int init_posix_timers(void)
 {
-	posix_timers_cache = kmem_cache_create("posix_timers_cache", sizeof(struct k_itimer), 0,
-					       SLAB_ACCOUNT, NULL);
+	posix_timers_cache = kmem_cache_create("posix_timers_cache", sizeof(struct k_itimer),
+					       __alignof__(struct k_itimer), SLAB_ACCOUNT, NULL);
 	return 0;
 }
 __initcall(init_posix_timers);

[patch V2 14/17] posix-timers: Make per process list RCU safe

[Thomas Gleixner]

Preparatory change to remove the sighand locking from the /proc/$PID/timers
iterator.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/time/posix-timers.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -510,7 +510,7 @@ static int do_timer_create(clockid_t whi
 		 * unmodified signal pointer to make it valid.
 		 */
 		WRITE_ONCE(new_timer->it_signal, current->signal);
-		hlist_add_head(&new_timer->list, &current->signal->posix_timers);
+		hlist_add_head_rcu(&new_timer->list, &current->signal->posix_timers);
 	}
 	/*
 	 * After unlocking sighand::siglock @new_timer is subject to
@@ -996,7 +996,7 @@ static void posix_timer_delete(struct k_
 		unsigned long sig = (unsigned long)timer->it_signal | 1UL;
 
 		WRITE_ONCE(timer->it_signal, (struct signal_struct *)sig);
-		hlist_del(&timer->list);
+		hlist_del_rcu(&timer->list);
 		posix_timer_cleanup_ignored(timer);
 	}
 

[patch V2 15/17] posix-timers: Dont iterate /proc/$PID/timers with sighand::siglock held

[Thomas Gleixner]

The readout of /proc/$PID/timers holds sighand::siglock with interrupts
disabled. That is required to protect against concurrent modifications of
the task::signal::posix_timers list because the list is not RCU safe.

With the conversion of the timer storage to a RCU protected hlist, this is
not longer required.

The only requirement is to protect the returned entry against a concurrent
free, which is trivial as the timers are RCU protected.

Removing the trylock of sighand::siglock is benign because the life time of
task_struct::signal is bound to the life time of the task_struct itself.

There are two scenarios where this matters:

  1) The process is life and not about to be checkpointed

  2) The process is stopped via ptrace for checkpointing

#1 is a racy snapshot of the armed timers and nothing can rely on it. It's
   not more than debug information and it has been that way before because
   sighand lock is dropped when the buffer is full and the restart of
   the iteration might find a completely different set of timers.

   The task and therefore task::signal cannot be freed as timers_start()
   acquired a reference count via get_pid_task().

#2 the process is stopped for checkpointing so nothing can delete or create
   timers at this point. Neither can the process exit during the traversal.

   If CRIU fails to observe an exit in progress prior to the dissimination
   of the timers, then there are more severe problems to solve in the CRIU
   mechanics as they can't rely on posix timers being enabled in the first
   place.

Therefore replace the lock acquisition with rcu_read_lock() and switch the
timer storage traversal over to seq_hlist_*_rcu().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 fs/proc/base.c |   48 ++++++++++++++++++++----------------------------
 1 file changed, 20 insertions(+), 28 deletions(-)

--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -2497,11 +2497,9 @@ static const struct file_operations proc
 
 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
 struct timers_private {
-	struct pid *pid;
-	struct task_struct *task;
-	struct sighand_struct *sighand;
-	struct pid_namespace *ns;
-	unsigned long flags;
+	struct pid		*pid;
+	struct task_struct	*task;
+	struct pid_namespace	*ns;
 };
 
 static void *timers_start(struct seq_file *m, loff_t *pos)
@@ -2512,54 +2510,48 @@ static void *timers_start(struct seq_fil
 	if (!tp->task)
 		return ERR_PTR(-ESRCH);
 
-	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
-	if (!tp->sighand)
-		return ERR_PTR(-ESRCH);
-
-	return seq_hlist_start(&tp->task->signal->posix_timers, *pos);
+	rcu_read_lock();
+	return seq_hlist_start_rcu(&tp->task->signal->posix_timers, *pos);
 }
 
 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
 {
 	struct timers_private *tp = m->private;
-	return seq_hlist_next(v, &tp->task->signal->posix_timers, pos);
+
+	return seq_hlist_next_rcu(v, &tp->task->signal->posix_timers, pos);
 }
 
 static void timers_stop(struct seq_file *m, void *v)
 {
 	struct timers_private *tp = m->private;
 
-	if (tp->sighand) {
-		unlock_task_sighand(tp->task, &tp->flags);
-		tp->sighand = NULL;
-	}
-
 	if (tp->task) {
 		put_task_struct(tp->task);
 		tp->task = NULL;
+		rcu_read_unlock();
 	}
 }
 
 static int show_timer(struct seq_file *m, void *v)
 {
-	struct k_itimer *timer;
-	struct timers_private *tp = m->private;
-	int notify;
 	static const char * const nstr[] = {
-		[SIGEV_SIGNAL] = "signal",
-		[SIGEV_NONE] = "none",
-		[SIGEV_THREAD] = "thread",
+		[SIGEV_SIGNAL]	= "signal",
+		[SIGEV_NONE]	= "none",
+		[SIGEV_THREAD]	= "thread",
 	};
 
-	timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
-	notify = timer->it_sigev_notify;
+	struct k_itimer *timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
+	struct timers_private *tp = m->private;
+	int notify = timer->it_sigev_notify;
+
+	guard(spinlock_irq)(&timer->it_lock);
+	if (!posixtimer_valid(timer))
+		return 0;
 
 	seq_printf(m, "ID: %d\n", timer->it_id);
-	seq_printf(m, "signal: %d/%px\n",
-		   timer->sigq.info.si_signo,
+	seq_printf(m, "signal: %d/%px\n", timer->sigq.info.si_signo,
 		   timer->sigq.info.si_value.sival_ptr);
-	seq_printf(m, "notify: %s/%s.%d\n",
-		   nstr[notify & ~SIGEV_THREAD_ID],
+	seq_printf(m, "notify: %s/%s.%d\n", nstr[notify & ~SIGEV_THREAD_ID],
 		   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
 		   pid_nr_ns(timer->it_pid, tp->ns));
 	seq_printf(m, "ClockID: %d\n", timer->it_clock);

[patch V2 16/17] posix-timers: Provide a mechanism to allocate a given timer ID

[Thomas Gleixner]

Checkpoint/Restore in Userspace (CRIU) requires to reconstruct posix timers
with the same timer ID on restore. It uses sys_timer_create() and relies on
the monotonic increasing timer ID provided by this syscall. It creates and
deletes timers until the desired ID is reached. This is can loop for a long
time, when the checkpointed process had a very sparse timer ID range.

It has been debated to implement a new syscall to allow the creation of
timers with a given timer ID, but that's tideous due to the 32/64bit compat
issues of sigevent_t and of dubious value.

The restore mechanism of CRIU creates the timers in a state where all
threads of the restored process are held on a barrier and cannot issue
syscalls. That means the restorer task has exclusive control.

This allows to address this issue with a prctl() so that the restorer
thread can do:

   if (prctl(PR_TIMER_CREATE_RESTORE_IDS, PR_TIMER_CREATE_RESTORE_IDS_ON))
      goto linear_mode;
   create_timers_with_explicit_ids();
   prctl(PR_TIMER_CREATE_RESTORE_IDS, PR_TIMER_CREATE_RESTORE_IDS_OFF);
   
This is backwards compatible because the prctl() fails on older kernels and
CRIU can fall back to the linear timer ID mechanism. CRIU versions which do
not know about the prctl() just work as before.

Implement the prctl() and modify timer_create() so that it copies the
requested timer ID from userspace by utilizing the existing timer_t
pointer, which is used to copy out the allocated timer ID on success.

If the prctl() is disabled, which it is by default, timer_create() works as
before and does not try to read from the userspace pointer.

There is no problem when a broken or rogue user space application enables
the prctl(). If the user space pointer does not contain a valid ID, then
timer_create() fails. If the data is not initialized, but constains a
random valid ID, timer_create() will create that random timer ID or fail if
the ID is already given out. 
 
As CRIU must use the raw syscall to avoid manipulating the internal state
of the restored process, this has no library dependencies and can be
adopted by CRIU right away.

Recreating two timers with IDs 1000000 and 2000000 takes 1.5 seconds with
the create/delete method. With the prctl() it takes 3 microseconds.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Move the ID counter ahead to avoid collisions after switching back to
    normal mode.
---
 include/linux/posix-timers.h |    2 
 include/linux/sched/signal.h |    1 
 include/uapi/linux/prctl.h   |   10 ++++
 kernel/sys.c                 |    5 ++
 kernel/time/posix-timers.c   |   97 +++++++++++++++++++++++++++++++------------
 5 files changed, 89 insertions(+), 26 deletions(-)

--- a/include/linux/posix-timers.h
+++ b/include/linux/posix-timers.h
@@ -114,6 +114,7 @@ bool posixtimer_init_sigqueue(struct sig
 void posixtimer_send_sigqueue(struct k_itimer *tmr);
 bool posixtimer_deliver_signal(struct kernel_siginfo *info, struct sigqueue *timer_sigq);
 void posixtimer_free_timer(struct k_itimer *timer);
+long posixtimer_create_prctl(unsigned long ctrl);
 
 /* Init task static initializer */
 #define INIT_CPU_TIMERBASE(b) {						\
@@ -140,6 +141,7 @@ static inline void posixtimer_rearm_itim
 static inline bool posixtimer_deliver_signal(struct kernel_siginfo *info,
 					     struct sigqueue *timer_sigq) { return false; }
 static inline void posixtimer_free_timer(struct k_itimer *timer) { }
+static inline long posixtimer_create_prctl(unsigned long ctrl) { return -EINVAL; }
 #endif
 
 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
--- a/include/linux/sched/signal.h
+++ b/include/linux/sched/signal.h
@@ -136,6 +136,7 @@ struct signal_struct {
 #ifdef CONFIG_POSIX_TIMERS
 
 	/* POSIX.1b Interval Timers */
+	unsigned int		timer_create_restore_ids:1;
 	atomic_t		next_posix_timer_id;
 	struct hlist_head	posix_timers;
 	struct hlist_head	ignored_posix_timers;
--- a/include/uapi/linux/prctl.h
+++ b/include/uapi/linux/prctl.h
@@ -353,4 +353,14 @@ struct prctl_mm_map {
  */
 #define PR_LOCK_SHADOW_STACK_STATUS      76
 
+/*
+ * Controls the mode of timer_create() for CRIU restore operations.
+ * Enabling this allows CRIU to restore timers with explicit IDs.
+ *
+ * Don't use for normal operations as the result might be undefined.
+ */
+#define PR_TIMER_CREATE_RESTORE_IDS		77
+# define PR_TIMER_CREATE_RESTORE_IDS_OFF	0
+# define PR_TIMER_CREATE_RESTORE_IDS_ON		1
+
 #endif /* _LINUX_PRCTL_H */
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -2811,6 +2811,11 @@ SYSCALL_DEFINE5(prctl, int, option, unsi
 			return -EINVAL;
 		error = arch_lock_shadow_stack_status(me, arg2);
 		break;
+	case PR_TIMER_CREATE_RESTORE_IDS:
+		if (arg3 || arg4 || arg5)
+			return -EINVAL;
+		error = posixtimer_create_prctl(arg2);
+		break;
 	default:
 		trace_task_prctl_unknown(option, arg2, arg3, arg4, arg5);
 		error = -EINVAL;
--- a/kernel/time/posix-timers.c
+++ b/kernel/time/posix-timers.c
@@ -19,6 +19,7 @@
 #include <linux/nospec.h>
 #include <linux/posix-clock.h>
 #include <linux/posix-timers.h>
+#include <linux/prctl.h>
 #include <linux/sched/task.h>
 #include <linux/slab.h>
 #include <linux/syscalls.h>
@@ -57,6 +58,8 @@ static const struct k_clock * const posi
 static const struct k_clock *clockid_to_kclock(const clockid_t id);
 static const struct k_clock clock_realtime, clock_monotonic;
 
+#define TIMER_ANY_ID		INT_MIN
+
 /* SIGEV_THREAD_ID cannot share a bit with the other SIGEV values. */
 #if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
 			~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
@@ -127,38 +130,60 @@ static bool posix_timer_hashed(struct ti
 	return false;
 }
 
-static int posix_timer_add(struct k_itimer *timer)
+static bool posix_timer_add_at(struct k_itimer *timer, struct signal_struct *sig, unsigned int id)
+{
+	struct timer_hash_bucket *bucket = hash_bucket(sig, id);
+
+	scoped_guard (spinlock, &bucket->lock) {
+		/*
+		 * Validate under the lock as this could have raced against
+		 * another thread ending up with the same ID, which is
+		 * highly unlikely, but possible.
+		 */
+		if (!posix_timer_hashed(bucket, sig, id)) {
+			/*
+			 * Set the timer ID and the signal pointer to make
+			 * it identifiable in the hash table. The signal
+			 * pointer has bit 0 set to indicate that it is not
+			 * yet fully initialized. posix_timer_hashed()
+			 * masks this bit out, but the syscall lookup fails
+			 * to match due to it being set. This guarantees
+			 * that there can't be duplicate timer IDs handed
+			 * out.
+			 */
+			timer->it_id = (timer_t)id;
+			timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
+			hlist_add_head_rcu(&timer->t_hash, &bucket->head);
+			return true;
+		}
+	}
+	return false;
+}
+
+static int posix_timer_add(struct k_itimer *timer, int req_id)
 {
 	struct signal_struct *sig = current->signal;
 
+	if (unlikely(req_id != TIMER_ANY_ID)) {
+		if (!posix_timer_add_at(timer, sig, req_id))
+			return -EBUSY;
+
+		/*
+		 * Move the ID counter past the requested ID, so that after
+		 * switching back to normal mode the IDs are outside of the
+		 * exact allocated region. That avoids ID collisions on the
+		 * next regular timer_create() invocations.
+		 */
+		atomic_set(&sig->next_posix_timer_id, req_id + 1);
+		return req_id;
+	}
+
 	for (unsigned int cnt = 0; cnt <= INT_MAX; cnt++) {
 		/* Get the next timer ID and clamp it to positive space */
 		unsigned int id = atomic_fetch_inc(&sig->next_posix_timer_id) & INT_MAX;
-		struct timer_hash_bucket *bucket = hash_bucket(sig, id);
 
-		scoped_guard (spinlock, &bucket->lock) {
-			/*
-			 * Validate under the lock as this could have raced
-			 * against another thread ending up with the same
-			 * ID, which is highly unlikely, but possible.
-			 */
-			if (!posix_timer_hashed(bucket, sig, id)) {
-				/*
-				 * Set the timer ID and the signal pointer to make
-				 * it identifiable in the hash table. The signal
-				 * pointer has bit 0 set to indicate that it is not
-				 * yet fully initialized. posix_timer_hashed()
-				 * masks this bit out, but the syscall lookup fails
-				 * to match due to it being set. This guarantees
-				 * that there can't be duplicate timer IDs handed
-				 * out.
-				 */
-				timer->it_id = (timer_t)id;
-				timer->it_signal = (struct signal_struct *)((unsigned long)sig | 1UL);
-				hlist_add_head_rcu(&timer->t_hash, &bucket->head);
-				return id;
-			}
-		}
+		if (posix_timer_add_at(timer, sig, id))
+			return id;
 		cond_resched();
 	}
 	/* POSIX return code when no timer ID could be allocated */
@@ -363,6 +388,16 @@ static enum hrtimer_restart posix_timer_
 	return HRTIMER_NORESTART;
 }
 
+long posixtimer_create_prctl(unsigned long ctrl)
+{
+	if (ctrl > PR_TIMER_CREATE_RESTORE_IDS_ON)
+		return -EINVAL;
+
+	guard(spinlock_irq)(&current->sighand->siglock);
+	current->signal->timer_create_restore_ids = ctrl == PR_TIMER_CREATE_RESTORE_IDS_ON;
+	return 0;
+}
+
 static struct pid *good_sigevent(sigevent_t * event)
 {
 	struct pid *pid = task_tgid(current);
@@ -434,6 +469,7 @@ static int do_timer_create(clockid_t whi
 			   timer_t __user *created_timer_id)
 {
 	const struct k_clock *kc = clockid_to_kclock(which_clock);
+	timer_t req_id = TIMER_ANY_ID;
 	struct k_itimer *new_timer;
 	int error, new_timer_id;
 
@@ -448,11 +484,20 @@ static int do_timer_create(clockid_t whi
 
 	spin_lock_init(&new_timer->it_lock);
 
+	/* Special case for CRIU to restore timers with a given timer ID. */
+	if (unlikely(current->signal->timer_create_restore_ids)) {
+		if (copy_from_user(&req_id, created_timer_id, sizeof(req_id)))
+			return -EFAULT;
+		/* Valid IDs are 0..INT_MAX */
+		if ((unsigned int)req_id > INT_MAX)
+			return -EINVAL;
+	}
+
 	/*
 	 * Add the timer to the hash table. The timer is not yet valid
 	 * after insertion, but has a unique ID allocated.
 	 */
-	new_timer_id = posix_timer_add(new_timer);
+	new_timer_id = posix_timer_add(new_timer, req_id);
 	if (new_timer_id < 0) {
 		posixtimer_free_timer(new_timer);
 		return new_timer_id;

[patch V2 17/17] selftests/timers/posix-timers: Add a test for exact allocation mode

[Thomas Gleixner]

The exact timer ID allocation mode is used by CRIU to restore timers with a
given ID. Add a test case for it.

It's skipped on older kernels when the prctl() fails.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Adopt to the ID counter change in the exact mode case
---
 tools/testing/selftests/timers/posix_timers.c |   60 +++++++++++++++++++++++++-
 1 file changed, 59 insertions(+), 1 deletion(-)

--- a/tools/testing/selftests/timers/posix_timers.c
+++ b/tools/testing/selftests/timers/posix_timers.c
@@ -7,6 +7,7 @@
  * Kernel loop code stolen from Steven Rostedt <srostedt@redhat.com>
  */
 #define _GNU_SOURCE
+#include <sys/prctl.h>
 #include <sys/time.h>
 #include <sys/types.h>
 #include <stdio.h>
@@ -599,14 +600,71 @@ static void check_overrun(int which, con
 			 "check_overrun %s\n", name);
 }
 
+#include <sys/syscall.h>
+
+static int do_timer_create(int *id)
+{
+	return syscall(__NR_timer_create, CLOCK_MONOTONIC, NULL, id);
+}
+
+static int do_timer_delete(int id)
+{
+	return syscall(__NR_timer_delete, id);
+}
+
+static void check_timer_create_exact(void)
+{
+	int id;
+
+	if (prctl(77, 1, 0, 0, 0)) {
+		switch (errno) {
+		case EINVAL:
+			ksft_test_result_skip("check timer create exact, not supported\n");
+			return;
+		default:
+			ksft_test_result_skip("check timer create exact, errno = %d\n", errno);
+			return;
+		}
+	}
+
+	id = 8;
+	if (do_timer_create(&id) < 0)
+		fatal_error(NULL, "timer_create()");
+
+	if (do_timer_delete(id))
+		fatal_error(NULL, "timer_delete()");
+
+	if (prctl(77, 0, 0, 0, 0))
+		fatal_error(NULL, "prctl()");
+
+	if (id != 8) {
+		ksft_test_result_fail("check timer create exact %d != 8\n", id);
+		return;
+	}
+
+	/* Validate that it went back to normal mode and allocates ID 9 */
+	if (do_timer_create(&id) < 0)
+		fatal_error(NULL, "timer_create()");
+
+	if (do_timer_delete(id))
+		fatal_error(NULL, "timer_delete()");
+
+	if (id == 9)
+		ksft_test_result_pass("check timer create exact\n");
+	else
+		ksft_test_result_fail("check timer create exact. Disabling failed.\n");
+}
+
 int main(int argc, char **argv)
 {
 	ksft_print_header();
-	ksft_set_plan(18);
+	ksft_set_plan(19);
 
 	ksft_print_msg("Testing posix timers. False negative may happen on CPU execution \n");
 	ksft_print_msg("based timers if other threads run on the CPU...\n");
 
+	check_timer_create_exact();
+
 	check_itimer(ITIMER_VIRTUAL, "ITIMER_VIRTUAL");
 	check_itimer(ITIMER_PROF, "ITIMER_PROF");
 	check_itimer(ITIMER_REAL, "ITIMER_REAL");