Skip to contentSkip to conversation

A couple of sched_getattr() fixes for DL

PATCH Applied Quentin Perret 4 participants 11 messages Jul 27, 2021 – Aug 5, 2021 Scheduler

lore
Share & tools
lore
#0 lore raw

cover A couple of sched_getattr() fixes for DL

Hi all,

In the context of [1] it became apparent that the way __getparam_dl() copies the sched_flags into the dl_entity struct for deadline tasks can have undesirable side effects. This series fixes two issues in that area causing sched_getattr() to report stale values or things it shouldn't report to userspace.

Thanks,
Quentin

[1] https://lore.kernel.org/lkml/[email protected]/

Quentin Perret (2):
sched/deadline: Fix reset_on_fork reporting of DL tasks
sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

 kernel/sched/core.c     | 1 +
 kernel/sched/deadline.c | 7 ++++---
 kernel/sched/sched.h    | 2 ++
 3 files changed, 7 insertions(+), 3 deletions(-)
Signature · 1 line
2.32.0.432.gabb21c7263-goog
#1 lore raw

PATCH 1/2 sched/deadline: Fix reset_on_fork reporting of DL tasks

It is possible for sched_getattr() to incorrectly report the state of the reset_on_fork flag when called on a deadline task.

Indeed, if the flag was set on a deadline task using sched_setattr() with flags (SCHED_FLAG_RESET_ON_FORK | SCHED_FLAG_KEEP_PARAMS), then p->sched_reset_on_fork will be set, but __setscheduler() will bail out early, which means that the dl_se->flags will not get updated by __setscheduler_params()->__setparam_dl(). Consequently, if sched_getattr() is then called on the task, __getparam_dl() will override kattr.sched_flags with the now out-of-date copy in dl_se->flags and report the stale value to userspace.

To fix this, make sure to only copy the flags that are relevant to sched_deadline to and from the dl_se->flags field.

---
 kernel/sched/deadline.c | 7 ++++---
 kernel/sched/sched.h    | 2 ++
 2 files changed, 6 insertions(+), 3 deletions(-)
kernel/sched/deadline.c +4 -3
index aaacd6cfd42f..5cafc642e647 100644
@@ -2741,7 +2741,7 @@ void __setparam_dl(struct task_struct *p, const struct sched_attr *attr)
 	dl_se->dl_runtime = attr->sched_runtime;
 	dl_se->dl_deadline = attr->sched_deadline;
 	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
-	dl_se->flags = attr->sched_flags;
+	dl_se->flags = attr->sched_flags & SCHED_DL_FLAGS;
 	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
 	dl_se->dl_density = to_ratio(dl_se->dl_deadline, dl_se->dl_runtime);
 }
@@ -2754,7 +2754,8 @@ void __getparam_dl(struct task_struct *p, struct sched_attr *attr)
 	attr->sched_runtime = dl_se->dl_runtime;
 	attr->sched_deadline = dl_se->dl_deadline;
 	attr->sched_period = dl_se->dl_period;
-	attr->sched_flags = dl_se->flags;
+	attr->sched_flags &= ~SCHED_DL_FLAGS;
+	attr->sched_flags |= dl_se->flags;
 }
 
 /*
@@ -2851,7 +2852,7 @@ bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr)
 	if (dl_se->dl_runtime != attr->sched_runtime ||
 	    dl_se->dl_deadline != attr->sched_deadline ||
 	    dl_se->dl_period != attr->sched_period ||
-	    dl_se->flags != attr->sched_flags)
+	    dl_se->flags != (attr->sched_flags & SCHED_DL_FLAGS))
 		return true;
 
 	return false;
kernel/sched/sched.h +2 -0
index 14a41a243f7b..ad3aee63db26 100644
@@ -227,6 +227,8 @@ static inline void update_avg(u64 *avg, u64 sample)
  */
 #define SCHED_FLAG_SUGOV	0x10000000
 
+#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
+
 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
 {
 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
Signature · 1 line
2.32.0.432.gabb21c7263-goog
reviewer Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] sched/deadline: Fix reset_on_fork reporting of DL tasks

Quoted 8 lines · Quentin Perret

It is possible for sched_getattr() to incorrectly report the state of the reset_on_fork flag when called on a deadline task.

Indeed, if the flag was set on a deadline task using sched_setattr() with flags (SCHED_FLAG_RESET_ON_FORK | SCHED_FLAG_KEEP_PARAMS), then p->sched_reset_on_fork will be set, but __setscheduler() will bail out early, which means that the dl_se->flags will not get updated by __setscheduler_params()->__setparam_dl(). Consequently, if

True, but it would also be awkward if non-DL related flags would have to be stored in dl_se->flags.

Quoted 6 lines

sched_getattr() is then called on the task, __getparam_dl() will override kattr.sched_flags with the now out-of-date copy in dl_se->flags and report the stale value to userspace.

To fix this, make sure to only copy the flags that are relevant to sched_deadline to and from the dl_se->flags field.

It also fixes the 'hidden' issue that a

uclampset -mX -MY -p dl_task

would end up at 'change:' label because of

    dl_se->flags != attr->sched_flags

and not because of

    attr->sched_flags & SCHED_FLAG_UTIL_CLAMP

And it also unblocks the uclamp-dl issue raised in
https://lkml.kernel.org/r/[email protected]
which surfaced when using get_params()->__getparam_dl() in SYSCALL_DEFINE3(sched_setattr,...).
Just for reference, IIRC, you mentioned this already on irc.

[...]

merge bot Applied
#3 lore raw

[tip: sched/core] sched/deadline: Fix reset_on_fork reporting of DL tasks

In reply to Quentin Perret
The following commit has been merged into the sched/core branch of tip:

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     f95091536f78971b269ec321b057b8d630b0ad8a
Gitweb:        https://git.kernel.org/tip/f95091536f78971b269ec321b057b8d630b0ad8a
Author:        Quentin Perret <[email protected]>
AuthorDate:    Tue, 27 Jul 2021 11:11:01 +01:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Wed, 04 Aug 2021 15:16:43 +02:00

sched/deadline: Fix reset_on_fork reporting of DL tasks

It is possible for sched_getattr() to incorrectly report the state of the reset_on_fork flag when called on a deadline task.

Indeed, if the flag was set on a deadline task using sched_setattr() with flags (SCHED_FLAG_RESET_ON_FORK | SCHED_FLAG_KEEP_PARAMS), then p->sched_reset_on_fork will be set, but __setscheduler() will bail out early, which means that the dl_se->flags will not get updated by __setscheduler_params()->__setparam_dl(). Consequently, if sched_getattr() is then called on the task, __getparam_dl() will override kattr.sched_flags with the now out-of-date copy in dl_se->flags and report the stale value to userspace.

To fix this, make sure to only copy the flags that are relevant to sched_deadline to and from the dl_se->flags field.

---
 kernel/sched/deadline.c | 7 ++++---
 kernel/sched/sched.h    | 2 ++
 2 files changed, 6 insertions(+), 3 deletions(-)
kernel/sched/deadline.c +4 -3
index aaacd6c..5cafc64 100644
@@ -2741,7 +2741,7 @@ void __setparam_dl(struct task_struct *p, const struct sched_attr *attr)
 	dl_se->dl_runtime = attr->sched_runtime;
 	dl_se->dl_deadline = attr->sched_deadline;
 	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
-	dl_se->flags = attr->sched_flags;
+	dl_se->flags = attr->sched_flags & SCHED_DL_FLAGS;
 	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
 	dl_se->dl_density = to_ratio(dl_se->dl_deadline, dl_se->dl_runtime);
 }
@@ -2754,7 +2754,8 @@ void __getparam_dl(struct task_struct *p, struct sched_attr *attr)
 	attr->sched_runtime = dl_se->dl_runtime;
 	attr->sched_deadline = dl_se->dl_deadline;
 	attr->sched_period = dl_se->dl_period;
-	attr->sched_flags = dl_se->flags;
+	attr->sched_flags &= ~SCHED_DL_FLAGS;
+	attr->sched_flags |= dl_se->flags;
 }
 
 /*
@@ -2851,7 +2852,7 @@ bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr)
 	if (dl_se->dl_runtime != attr->sched_runtime ||
 	    dl_se->dl_deadline != attr->sched_deadline ||
 	    dl_se->dl_period != attr->sched_period ||
-	    dl_se->flags != attr->sched_flags)
+	    dl_se->flags != (attr->sched_flags & SCHED_DL_FLAGS))
 		return true;
 
 	return false;
kernel/sched/sched.h +2 -0
index 9a1c6ae..75a5c12 100644
@@ -227,6 +227,8 @@ static inline void update_avg(u64 *avg, u64 sample)
  */
 #define SCHED_FLAG_SUGOV	0x10000000
 
+#define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
+
 static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se)
 {
 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
#4 lore raw

PATCH 2/2 sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

In reply to Quentin Perret

SCHED_FLAG_SUGOV is supposed to be a kernel-only flag that userspace cannot interact with. However, sched_getattr() currently reports it in sched_flags if called on a sugov worker even though it is not actually defined in a UAPI header. To avoid this, make sure to clean-up the sched_flags field in sched_getattr() before returning to userspace.

---
 kernel/sched/core.c | 1 +
 1 file changed, 1 insertion(+)
kernel/sched/core.c +1 -0
index 2d9ff40f4661..d8f489dcc383 100644
@@ -7535,6 +7535,7 @@ SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
 		kattr.sched_priority = p->rt_priority;
 	else
 		kattr.sched_nice = task_nice(p);
+	kattr.sched_flags &= SCHED_FLAG_ALL;
 
 #ifdef CONFIG_UCLAMP_TASK
 	/*
Signature · 1 line
2.32.0.432.gabb21c7263-goog
maintainer
#5 lore raw

Re: [PATCH 2/2] sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

Hi Quentin,

Quoted 17 lines · Quentin Perret

SCHED_FLAG_SUGOV is supposed to be a kernel-only flag that userspace cannot interact with. However, sched_getattr() currently reports it in sched_flags if called on a sugov worker even though it is not actually defined in a UAPI header. To avoid this, make sure to clean-up the sched_flags field in sched_getattr() before returning to userspace.

---
 kernel/sched/core.c | 1 +
 1 file changed, 1 insertion(+)
kernel/sched/core.c +0 -0
index 2d9ff40f4661..d8f489dcc383 100644

kernel/sched/core.c SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, · line 7538

Quoted 4 lines
kernel/sched/core.c +1 -0
@@ -7535,6 +7535,7 @@ SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
 		kattr.sched_priority = p->rt_priority;
 	else
 		kattr.sched_nice = task_nice(p);
+	kattr.sched_flags &= SCHED_FLAG_ALL;

Maybe we can do this in the previous patch so that it's kept confined to deadline bits?

Thanks,
Juri

#6 lore raw

Re: [PATCH 2/2] sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

Quoted 15 lines · Juri Lelli

Hi Quentin,

Quoted 12 lines · Quentin Perret

SCHED_FLAG_SUGOV is supposed to be a kernel-only flag that userspace cannot interact with. However, sched_getattr() currently reports it in sched_flags if called on a sugov worker even though it is not actually defined in a UAPI header. To avoid this, make sure to clean-up the sched_flags field in sched_getattr() before returning to userspace.

---
 kernel/sched/core.c | 1 +
 1 file changed, 1 insertion(+)
Quoted 12 lines

kernel/sched/core.c SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, · line 7538

Quoted 9 lines
kernel/sched/core.c +1 -0
index 2d9ff40f4661..d8f489dcc383 100644
@@ -7535,6 +7535,7 @@ SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
 		kattr.sched_priority = p->rt_priority;
 	else
 		kattr.sched_nice = task_nice(p);
+	kattr.sched_flags &= SCHED_FLAG_ALL;

Maybe we can do this in the previous patch so that it's kept confined to deadline bits?

That works too, it just felt like this could happen again if we start using non-standard flags outside of deadline for any reason at some point in the future. But no strong opinion really.

Cheers,
Quentin

maintainer
#7 lore raw

Re: [PATCH 2/2] sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

Quoted 29 lines · Quentin Perret
Quoted 15 lines · Juri Lelli

Hi Quentin,

Quoted 12 lines · Quentin Perret

SCHED_FLAG_SUGOV is supposed to be a kernel-only flag that userspace cannot interact with. However, sched_getattr() currently reports it in sched_flags if called on a sugov worker even though it is not actually defined in a UAPI header. To avoid this, make sure to clean-up the sched_flags field in sched_getattr() before returning to userspace.

---
 kernel/sched/core.c | 1 +
 1 file changed, 1 insertion(+)
Quoted 12 lines

kernel/sched/core.c SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, · line 7538

Quoted 9 lines
kernel/sched/core.c +1 -0
index 2d9ff40f4661..d8f489dcc383 100644
@@ -7535,6 +7535,7 @@ SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
 		kattr.sched_priority = p->rt_priority;
 	else
 		kattr.sched_nice = task_nice(p);
+	kattr.sched_flags &= SCHED_FLAG_ALL;

Maybe we can do this in the previous patch so that it's kept confined to deadline bits?

Quoted 3 lines

That works too, it just felt like this could happen again if we start using non-standard flags outside of deadline for any reason at some point in the future. But no strong opinion really.

Yeah, I also see this point. :)

So no prob with me to keep it in core.c as you do here.

Best,
Juri

#8 lore raw

Re: [PATCH 2/2] sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

Quoted 2 lines · Juri Lelli
Quoted 1 line · Quentin Perret

On Wednesday 28 Jul 2021 at 11:12:03 (+0200), Juri Lelli wrote:

[...]

Quoted 7 lines
Quoted 6 lines
Quoted 2 lines

Maybe we can do this in the previous patch so that it's kept confined to deadline bits?

That works too, it just felt like this could happen again if we start using non-standard flags outside of deadline for any reason at some point in the future. But no strong opinion really.

Quoted 7 lines

Yeah, I also see this point. :)

So no prob with me to keep it in core.c as you do here.

Best,
Juri

I would vote for not exporting SCHED_FLAG_SUGOV from __getparam_dl() in patch 1/2 to underpin the idea that this flag is a hack.

@ -2759,7 +2759,7 @@ void __getparam_dl(struct task_struct *p, struct sched_attr *attr)

        attr->sched_deadline = dl_se->dl_deadline;
        attr->sched_period = dl_se->dl_period;
        attr->sched_flags &= ~SCHED_DL_FLAGS;
-       attr->sched_flags |= dl_se->flags;
+       attr->sched_flags |= dl_se->flags & ~SCHED_FLAG_SUGOV;
#9 lore raw

Re: [PATCH 2/2] sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

Quoted 29 lines · Dietmar Eggemann
Quoted 2 lines · Juri Lelli
Quoted 1 line · Quentin Perret

On Wednesday 28 Jul 2021 at 11:12:03 (+0200), Juri Lelli wrote:

[...]

Quoted 7 lines
Quoted 6 lines
Quoted 2 lines

Maybe we can do this in the previous patch so that it's kept confined to deadline bits?

That works too, it just felt like this could happen again if we start using non-standard flags outside of deadline for any reason at some point in the future. But no strong opinion really.

Quoted 7 lines

Yeah, I also see this point. :)

So no prob with me to keep it in core.c as you do here.

Best,
Juri

I would vote for not exporting SCHED_FLAG_SUGOV from __getparam_dl() in patch 1/2 to underpin the idea that this flag is a hack.

@ -2759,7 +2759,7 @@ void __getparam_dl(struct task_struct *p, struct sched_attr *attr)

        attr->sched_deadline = dl_se->dl_deadline;
        attr->sched_period = dl_se->dl_period;
        attr->sched_flags &= ~SCHED_DL_FLAGS;

Review

Quoted 2 lines
(quoted diff) +1 -1
-       attr->sched_flags |= dl_se->flags;
+       attr->sched_flags |= dl_se->flags & ~SCHED_FLAG_SUGOV;

Alright, that's 2 votes against 1, you win!
I'll post a v2 shortly.

Cheers,
Quentin

merge bot Applied
#10 lore raw

[tip: sched/core] sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

In reply to Quentin Perret
The following commit has been merged into the sched/core branch of tip:

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     7ad721bf10718a4e480a27ded8bb16b8f6feb2d1
Gitweb:        https://git.kernel.org/tip/7ad721bf10718a4e480a27ded8bb16b8f6feb2d1
Author:        Quentin Perret <[email protected]>
AuthorDate:    Tue, 27 Jul 2021 11:11:02 +01:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Wed, 04 Aug 2021 15:16:44 +02:00

sched: Don't report SCHED_FLAG_SUGOV in sched_getattr()

SCHED_FLAG_SUGOV is supposed to be a kernel-only flag that userspace cannot interact with. However, sched_getattr() currently reports it in sched_flags if called on a sugov worker even though it is not actually defined in a UAPI header. To avoid this, make sure to clean-up the sched_flags field in sched_getattr() before returning to userspace.

---
 kernel/sched/core.c | 1 +
 1 file changed, 1 insertion(+)
kernel/sched/core.c +1 -0
index 6c562ad..314f70d 100644
@@ -7530,6 +7530,7 @@ SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
 		kattr.sched_priority = p->rt_priority;
 	else
 		kattr.sched_nice = task_nice(p);
+	kattr.sched_flags &= SCHED_FLAG_ALL;
 
 #ifdef CONFIG_UCLAMP_TASK
 	/*