The FUTEX_LOCK_PI futex operand uses a CLOCK_REALTIME based absolute
timeout since it was implemented, but it does not require that the
FUTEX_CLOCK_REALTIME flag is set, because that was introduced later.
In theory as none of the user space implementations can set the
FUTEX_CLOCK_REALTIME flag on this operand, it would be possible to
creatively abuse it and make the meaning invers, i.e. select CLOCK_REALTIME
when not set and CLOCK_MONOTONIC when set. But that's a nasty hackery.
Another option would be to have a new FUTEX_CLOCK_MONOTONIC flag only for
FUTEX_LOCK_PI, but that's also awkward because it does not allow libraries
to handle the timeout clock selection consistently.
So provide a new FUTEX_LOCK_PI2 operand which implements the timeout
semantics which the other operands use and leave FUTEX_LOCK_PI alone.
Reported-by: Kurt Kanzenbach <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
---
include/uapi/linux/futex.h | 1 +
kernel/futex.c | 6 +++++-
2 files changed, 6 insertions(+), 1 deletion(-)
--- a/include/uapi/linux/futex.h
+++ b/include/uapi/linux/futex.h
@@ -21,6 +21,7 @@
#define FUTEX_WAKE_BITSET 10
#define FUTEX_WAIT_REQUEUE_PI 11
#define FUTEX_CMP_REQUEUE_PI 12
+#define FUTEX_LOCK_PI2 13
#define FUTEX_PRIVATE_FLAG 128
#define FUTEX_CLOCK_REALTIME 256
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -3711,7 +3711,8 @@ long do_futex(u32 __user *uaddr, int op,
if (op & FUTEX_CLOCK_REALTIME) {
flags |= FLAGS_CLOCKRT;
- if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
+ if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI &&
+ cmd != FUTEX_LOCK_PI2)
return -ENOSYS;
}
@@ -3744,6 +3745,8 @@ long do_futex(u32 __user *uaddr, int op,
return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
case FUTEX_LOCK_PI:
flags |= FLAGS_CLOCKRT;
+ fallthrough;
+ case FUTEX_LOCK_PI2:
return futex_lock_pi(uaddr, flags, timeout, 0);
case FUTEX_UNLOCK_PI:
return futex_unlock_pi(uaddr, flags);
@@ -3764,6 +3767,7 @@ static inline bool futex_cmd_has_timeout
switch (cmd) {
case FUTEX_WAIT:
case FUTEX_LOCK_PI:
+ case FUTEX_LOCK_PI2:
case FUTEX_WAIT_BITSET:
case FUTEX_WAIT_REQUEUE_PI:
return true;
Hi Thomas,
Às 16:44 de 22/04/21, Thomas Gleixner escreveu:
> The FUTEX_LOCK_PI futex operand uses a CLOCK_REALTIME based absolute
> timeout since it was implemented, but it does not require that the
> FUTEX_CLOCK_REALTIME flag is set, because that was introduced later.
>
> In theory as none of the user space implementations can set the
> FUTEX_CLOCK_REALTIME flag on this operand, it would be possible to
> creatively abuse it and make the meaning invers, i.e. select CLOCK_REALTIME
> when not set and CLOCK_MONOTONIC when set. But that's a nasty hackery.
>
> Another option would be to have a new FUTEX_CLOCK_MONOTONIC flag only for
> FUTEX_LOCK_PI, but that's also awkward because it does not allow libraries
> to handle the timeout clock selection consistently.
>
> So provide a new FUTEX_LOCK_PI2 operand which implements the timeout
> semantics which the other operands use and leave FUTEX_LOCK_PI alone.
>
> Reported-by: Kurt Kanzenbach <[email protected]>
> Signed-off-by: Thomas Gleixner <[email protected]>
> ---
> include/uapi/linux/futex.h | 1 +
> kernel/futex.c | 6 +++++-
> 2 files changed, 6 insertions(+), 1 deletion(-)
>
> --- a/include/uapi/linux/futex.h
> +++ b/include/uapi/linux/futex.h
> @@ -21,6 +21,7 @@
> #define FUTEX_WAKE_BITSET 10
> #define FUTEX_WAIT_REQUEUE_PI 11
> #define FUTEX_CMP_REQUEUE_PI 12
> +#define FUTEX_LOCK_PI2 13
>
> #define FUTEX_PRIVATE_FLAG 128
> #define FUTEX_CLOCK_REALTIME 256
To keep consistency with other operations, maybe add a
FUTEX_LOCK_PI2_PRIVATE?
> --- a/kernel/futex.c
> +++ b/kernel/futex.c
> @@ -3711,7 +3711,8 @@ long do_futex(u32 __user *uaddr, int op,
>
> if (op & FUTEX_CLOCK_REALTIME) {
> flags |= FLAGS_CLOCKRT;
> - if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
> + if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI &&
> + cmd != FUTEX_LOCK_PI2)
> return -ENOSYS;
> }
As FUTEX_LOCK_PI, FUTEX_LOCK_PI2 also requires FUTEX_CMPXCHG right?
Then, add it here:
switch (cmd) {
case FUTEX_LOCK_PI:
+ case FUTEX_LOCK_PI2:
case FUTEX_UNLOCK_PI:
case FUTEX_TRYLOCK_PI:
case FUTEX_WAIT_REQUEUE_PI:
case FUTEX_CMP_REQUEUE_PI:
if (!futex_cmpxchg_enabled)
return -ENOSYS;
}
>
> @@ -3744,6 +3745,8 @@ long do_futex(u32 __user *uaddr, int op,
> return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
> case FUTEX_LOCK_PI:
> flags |= FLAGS_CLOCKRT;
> + fallthrough;
> + case FUTEX_LOCK_PI2:
> return futex_lock_pi(uaddr, flags, timeout, 0);
> case FUTEX_UNLOCK_PI:
> return futex_unlock_pi(uaddr, flags);
> @@ -3764,6 +3767,7 @@ static inline bool futex_cmd_has_timeout
> switch (cmd) {
> case FUTEX_WAIT:
> case FUTEX_LOCK_PI:
> + case FUTEX_LOCK_PI2:
> case FUTEX_WAIT_BITSET:
> case FUTEX_WAIT_REQUEUE_PI:
> return true;
>
Thanks,
André
André!
On Fri, Apr 23 2021 at 19:20, André Almeida wrote:
>> @@ -21,6 +21,7 @@
>> #define FUTEX_WAKE_BITSET 10
>> #define FUTEX_WAIT_REQUEUE_PI 11
>> #define FUTEX_CMP_REQUEUE_PI 12
>> +#define FUTEX_LOCK_PI2 13
>>
>> #define FUTEX_PRIVATE_FLAG 128
>> #define FUTEX_CLOCK_REALTIME 256
>
> To keep consistency with other operations, maybe add a
> FUTEX_LOCK_PI2_PRIVATE?
Good point! Missed that.
>> --- a/kernel/futex.c
>> +++ b/kernel/futex.c
>> @@ -3711,7 +3711,8 @@ long do_futex(u32 __user *uaddr, int op,
>>
>> if (op & FUTEX_CLOCK_REALTIME) {
>> flags |= FLAGS_CLOCKRT;
>> - if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
>> + if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI &&
>> + cmd != FUTEX_LOCK_PI2)
>> return -ENOSYS;
>> }
>
> As FUTEX_LOCK_PI, FUTEX_LOCK_PI2 also requires FUTEX_CMPXCHG right?
> Then, add it here:
>
> switch (cmd) {
> case FUTEX_LOCK_PI:
> + case FUTEX_LOCK_PI2:
> case FUTEX_UNLOCK_PI:
> case FUTEX_TRYLOCK_PI:
> case FUTEX_WAIT_REQUEUE_PI:
> case FUTEX_CMP_REQUEUE_PI:
> if (!futex_cmpxchg_enabled)
> return -ENOSYS;
> }
Indeed. Forgot about that completely.
Thanks for spotting that!
tglx
The following commit has been merged into the locking/core branch of tip:
Commit-ID: bf22a6976897977b0a3f1aeba6823c959fc4fdae
Gitweb: https://git.kernel.org/tip/bf22a6976897977b0a3f1aeba6823c959fc4fdae
Author: Thomas Gleixner <[email protected]>
AuthorDate: Thu, 22 Apr 2021 21:44:23 +02:00
Committer: Peter Zijlstra <[email protected]>
CommitterDate: Tue, 22 Jun 2021 16:42:09 +02:00
futex: Provide FUTEX_LOCK_PI2 to support clock selection
The FUTEX_LOCK_PI futex operand uses a CLOCK_REALTIME based absolute
timeout since it was implemented, but it does not require that the
FUTEX_CLOCK_REALTIME flag is set, because that was introduced later.
In theory as none of the user space implementations can set the
FUTEX_CLOCK_REALTIME flag on this operand, it would be possible to
creatively abuse it and make the meaning invers, i.e. select CLOCK_REALTIME
when not set and CLOCK_MONOTONIC when set. But that's a nasty hackery.
Another option would be to have a new FUTEX_CLOCK_MONOTONIC flag only for
FUTEX_LOCK_PI, but that's also awkward because it does not allow libraries
to handle the timeout clock selection consistently.
So provide a new FUTEX_LOCK_PI2 operand which implements the timeout
semantics which the other operands use and leave FUTEX_LOCK_PI alone.
Reported-by: Kurt Kanzenbach <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
---
include/uapi/linux/futex.h | 2 ++
kernel/futex.c | 7 ++++++-
2 files changed, 8 insertions(+), 1 deletion(-)
diff --git a/include/uapi/linux/futex.h b/include/uapi/linux/futex.h
index a89eb0a..235e5b2 100644
--- a/include/uapi/linux/futex.h
+++ b/include/uapi/linux/futex.h
@@ -21,6 +21,7 @@
#define FUTEX_WAKE_BITSET 10
#define FUTEX_WAIT_REQUEUE_PI 11
#define FUTEX_CMP_REQUEUE_PI 12
+#define FUTEX_LOCK_PI2 13
#define FUTEX_PRIVATE_FLAG 128
#define FUTEX_CLOCK_REALTIME 256
@@ -32,6 +33,7 @@
#define FUTEX_CMP_REQUEUE_PRIVATE (FUTEX_CMP_REQUEUE | FUTEX_PRIVATE_FLAG)
#define FUTEX_WAKE_OP_PRIVATE (FUTEX_WAKE_OP | FUTEX_PRIVATE_FLAG)
#define FUTEX_LOCK_PI_PRIVATE (FUTEX_LOCK_PI | FUTEX_PRIVATE_FLAG)
+#define FUTEX_LOCK_PI2_PRIVATE (FUTEX_LOCK_PI2 | FUTEX_PRIVATE_FLAG)
#define FUTEX_UNLOCK_PI_PRIVATE (FUTEX_UNLOCK_PI | FUTEX_PRIVATE_FLAG)
#define FUTEX_TRYLOCK_PI_PRIVATE (FUTEX_TRYLOCK_PI | FUTEX_PRIVATE_FLAG)
#define FUTEX_WAIT_BITSET_PRIVATE (FUTEX_WAIT_BITSET | FUTEX_PRIVATE_FLAG)
diff --git a/kernel/futex.c b/kernel/futex.c
index f820439..f832b64 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -3707,12 +3707,14 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
if (op & FUTEX_CLOCK_REALTIME) {
flags |= FLAGS_CLOCKRT;
- if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
+ if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI &&
+ cmd != FUTEX_LOCK_PI2)
return -ENOSYS;
}
switch (cmd) {
case FUTEX_LOCK_PI:
+ case FUTEX_LOCK_PI2:
case FUTEX_UNLOCK_PI:
case FUTEX_TRYLOCK_PI:
case FUTEX_WAIT_REQUEUE_PI:
@@ -3740,6 +3742,8 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
case FUTEX_LOCK_PI:
flags |= FLAGS_CLOCKRT;
+ fallthrough;
+ case FUTEX_LOCK_PI2:
return futex_lock_pi(uaddr, flags, timeout, 0);
case FUTEX_UNLOCK_PI:
return futex_unlock_pi(uaddr, flags);
@@ -3760,6 +3764,7 @@ static __always_inline bool futex_cmd_has_timeout(u32 cmd)
switch (cmd) {
case FUTEX_WAIT:
case FUTEX_LOCK_PI:
+ case FUTEX_LOCK_PI2:
case FUTEX_WAIT_BITSET:
case FUTEX_WAIT_REQUEUE_PI:
return true;