diff mbox series

powerpc/entry: Clear TIF_SYSCALL_RET before syscall error return

Message ID 20260828053811.1042300-1-sshegde@linux.ibm.com (mailing list archive)
State Under Review
Headers show
Series powerpc/entry: Clear TIF_SYSCALL_RET before syscall error return | expand

Commit Message

Shrikanth Hegde Aug. 28, 2026, 5:38 a.m. UTC
Shivaprasad reported a boot failure due to userspace processes crash on
abort() from libc.so.6. It was bisected to merge request
commit '3424d8c18a7d ("Merge tag  'core-entry-2026-08-17' of 
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")'

Upon checking the merge, when syscall_enter_from_user_mode_randomize_stack
fails, which could happen when a tracer like seccomp or ptrace intercepts
and skips the syscall, the code returns to userspace immediately without
clearing the intermediate flag which was set.

When the next syscall is made, it immediately aborts the valid syscall
since the flag is still set. Hence clear the flag on occurrence of first
failure.

Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
Closes: https://lore.kernel.org/all/e301014d-568f-4ed5-bc64-b8a85ca0b1e1@linux.ibm.com/
Fixes: 3424d8c18a7d ("Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
PS: I have kept the block below since earlier code was checking it
regardless of result of syscall_enter_from_user_mode. If it turns out
to be a redundant, it can be removed later.

 arch/powerpc/kernel/syscall.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Comments

Shivaprasad G Bhat Aug. 29, 2026, 6:01 a.m. UTC | #1
On 8/28/26 11:08 AM, Shrikanth Hegde wrote:
> Shivaprasad reported a boot failure due to userspace processes crash on
> abort() from libc.so.6. It was bisected to merge request
> commit '3424d8c18a7d ("Merge tag  'core-entry-2026-08-17' of
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")'
> 
> Upon checking the merge, when syscall_enter_from_user_mode_randomize_stack
> fails, which could happen when a tracer like seccomp or ptrace intercepts
> and skips the syscall, the code returns to userspace immediately without
> clearing the intermediate flag which was set.
> 
> When the next syscall is made, it immediately aborts the valid syscall
> since the flag is still set. Hence clear the flag on occurrence of first
> failure.
> 
> Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
> Closes: https://lore.kernel.org/all/e301014d-568f-4ed5-bc64-b8a85ca0b1e1@linux.ibm.com/
> Fixes: 3424d8c18a7d ("Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
> Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>

Tested-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>

Thanks,
Shivaprasad

> ---
> PS: I have kept the block below since earlier code was checking it
> regardless of result of syscall_enter_from_user_mode. If it turns out
> to be a redundant, it can be removed later.
> 
>   arch/powerpc/kernel/syscall.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
> index 4916c205c4bb..fbefe1927b10 100644
> --- a/arch/powerpc/kernel/syscall.c
> +++ b/arch/powerpc/kernel/syscall.c
> @@ -18,8 +18,10 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
>   	long ret;
>   	syscall_fn f;
>   
> -	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
> +	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
> +		clear_thread_flag(TIF_SYSCALL_RET);
>   		return syscall_get_error(current, regs);
> +	}
>   
>   	if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
>   		return syscall_get_error(current, regs);
Mukesh Kumar Chaurasiya Aug. 31, 2026, 4:39 a.m. UTC | #2
On Fri, Aug 28, 2026 at 11:08:11AM +0530, Shrikanth Hegde wrote:
> Shivaprasad reported a boot failure due to userspace processes crash on
> abort() from libc.so.6. It was bisected to merge request
> commit '3424d8c18a7d ("Merge tag  'core-entry-2026-08-17' of 
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")'
> 
> Upon checking the merge, when syscall_enter_from_user_mode_randomize_stack
> fails, which could happen when a tracer like seccomp or ptrace intercepts
> and skips the syscall, the code returns to userspace immediately without
> clearing the intermediate flag which was set.
> 
> When the next syscall is made, it immediately aborts the valid syscall
> since the flag is still set. Hence clear the flag on occurrence of first
> failure.
> 
> Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
> Closes: https://lore.kernel.org/all/e301014d-568f-4ed5-bc64-b8a85ca0b1e1@linux.ibm.com/
> Fixes: 3424d8c18a7d ("Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
> Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
> ---
> PS: I have kept the block below since earlier code was checking it
> regardless of result of syscall_enter_from_user_mode. If it turns out
> to be a redundant, it can be removed later.
> 
I think it is redundant as of now. The TIF_SYSCALL_RET flag is set when
error is set by the ptrace or seccomp, If the error value is set then
the syscall_enter_from_user_mode_randomize_stack will return false.
Hence the next check will become redundant. 

I also see that TIF_SYSCALL_RET is also set when processing the ptrace
syscall, Which can leave the flag set for next syscall execution. Which
can again trigger the same issue.

Regards,
Mukesh
>  arch/powerpc/kernel/syscall.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
> index 4916c205c4bb..fbefe1927b10 100644
> --- a/arch/powerpc/kernel/syscall.c
> +++ b/arch/powerpc/kernel/syscall.c
> @@ -18,8 +18,10 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
>  	long ret;
>  	syscall_fn f;
>  
> -	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
> +	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
> +		clear_thread_flag(TIF_SYSCALL_RET);
>  		return syscall_get_error(current, regs);
> +	}
>  
>  	if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
>  		return syscall_get_error(current, regs);
> -- 
> 2.47.3
>
Shrikanth Hegde Aug. 31, 2026, 9:04 a.m. UTC | #3
On 8/31/26 10:09 AM, Mukesh Kumar Chaurasiya wrote:
> On Fri, Aug 28, 2026 at 11:08:11AM +0530, Shrikanth Hegde wrote:
>> Shivaprasad reported a boot failure due to userspace processes crash on
>> abort() from libc.so.6. It was bisected to merge request
>> commit '3424d8c18a7d ("Merge tag  'core-entry-2026-08-17' of
>> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")'
>>
>> Upon checking the merge, when syscall_enter_from_user_mode_randomize_stack
>> fails, which could happen when a tracer like seccomp or ptrace intercepts
>> and skips the syscall, the code returns to userspace immediately without
>> clearing the intermediate flag which was set.
>>
>> When the next syscall is made, it immediately aborts the valid syscall
>> since the flag is still set. Hence clear the flag on occurrence of first
>> failure.
>>
>> Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
>> Closes: https://lore.kernel.org/all/e301014d-568f-4ed5-bc64-b8a85ca0b1e1@linux.ibm.com/
>> Fixes: 3424d8c18a7d ("Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
>> Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
>> ---
>> PS: I have kept the block below since earlier code was checking it
>> regardless of result of syscall_enter_from_user_mode. If it turns out
>> to be a redundant, it can be removed later.
>>
> I think it is redundant as of now. The TIF_SYSCALL_RET flag is set when
> error is set by the ptrace or seccomp, If the error value is set then
> the syscall_enter_from_user_mode_randomize_stack will return false.
> Hence the next check will become redundant.
> 
> I also see that TIF_SYSCALL_RET is also set when processing the ptrace
> syscall, Which can leave the flag set for next syscall execution. Which
> can again trigger the same issue.
> 

I believe it is a pre-existing bug which can be fixed independent of this issue.
Since merge request broke the userspace it is better we fix it right away and
this pre-existing issue fix can be done subsequently.

No?

> Regards,
> Mukesh
>>   arch/powerpc/kernel/syscall.c | 4 +++-
>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
>> index 4916c205c4bb..fbefe1927b10 100644
>> --- a/arch/powerpc/kernel/syscall.c
>> +++ b/arch/powerpc/kernel/syscall.c
>> @@ -18,8 +18,10 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
>>   	long ret;
>>   	syscall_fn f;
>>   
>> -	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
>> +	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
>> +		clear_thread_flag(TIF_SYSCALL_RET);
>>   		return syscall_get_error(current, regs);
>> +	}
>>   
>>   	if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
>>   		return syscall_get_error(current, regs);
>> -- 
>> 2.47.3
>>
Mukesh Kumar Chaurasiya Sept. 1, 2026, 7:13 a.m. UTC | #4
On Mon, Aug 31, 2026 at 02:34:10PM +0530, Shrikanth Hegde wrote:
> 
> 
> On 8/31/26 10:09 AM, Mukesh Kumar Chaurasiya wrote:
> > On Fri, Aug 28, 2026 at 11:08:11AM +0530, Shrikanth Hegde wrote:
> > > Shivaprasad reported a boot failure due to userspace processes crash on
> > > abort() from libc.so.6. It was bisected to merge request
> > > commit '3424d8c18a7d ("Merge tag  'core-entry-2026-08-17' of
> > > git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")'
> > > 
> > > Upon checking the merge, when syscall_enter_from_user_mode_randomize_stack
> > > fails, which could happen when a tracer like seccomp or ptrace intercepts
> > > and skips the syscall, the code returns to userspace immediately without
> > > clearing the intermediate flag which was set.
> > > 
> > > When the next syscall is made, it immediately aborts the valid syscall
> > > since the flag is still set. Hence clear the flag on occurrence of first
> > > failure.
> > > 
> > > Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
> > > Closes: https://lore.kernel.org/all/e301014d-568f-4ed5-bc64-b8a85ca0b1e1@linux.ibm.com/
> > > Fixes: 3424d8c18a7d ("Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
> > > Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
> > > ---
> > > PS: I have kept the block below since earlier code was checking it
> > > regardless of result of syscall_enter_from_user_mode. If it turns out
> > > to be a redundant, it can be removed later.
> > > 
> > I think it is redundant as of now. The TIF_SYSCALL_RET flag is set when
> > error is set by the ptrace or seccomp, If the error value is set then
> > the syscall_enter_from_user_mode_randomize_stack will return false.
> > Hence the next check will become redundant.
> > 
> > I also see that TIF_SYSCALL_RET is also set when processing the ptrace
> > syscall, Which can leave the flag set for next syscall execution. Which
> > can again trigger the same issue.
> > 
> 
> I believe it is a pre-existing bug which can be fixed independent of this issue.
> Since merge request broke the userspace it is better we fix it right away and
> this pre-existing issue fix can be done subsequently.
> 
> No?
> 
Agreed.
Sure we can send that as a separate fix.
With that,

Reviewed-by: Mukesh Kumar Chaurasiya (IBM) <mkchauras@gmail.com>

Regards
Mukesh.
> > Regards,
> > Mukesh
> > >   arch/powerpc/kernel/syscall.c | 4 +++-
> > >   1 file changed, 3 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
> > > index 4916c205c4bb..fbefe1927b10 100644
> > > --- a/arch/powerpc/kernel/syscall.c
> > > +++ b/arch/powerpc/kernel/syscall.c
> > > @@ -18,8 +18,10 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
> > >   	long ret;
> > >   	syscall_fn f;
> > > -	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
> > > +	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
> > > +		clear_thread_flag(TIF_SYSCALL_RET);
> > >   		return syscall_get_error(current, regs);
> > > +	}
> > >   	if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
> > >   		return syscall_get_error(current, regs);
> > > -- 
> > > 2.47.3
> > > 
>
Venkat Sept. 1, 2026, 9:17 a.m. UTC | #5
> On 28 Aug 2026, at 11:08 AM, Shrikanth Hegde <sshegde@linux.ibm.com> wrote:
> 
> Shivaprasad reported a boot failure due to userspace processes crash on
> abort() from libc.so.6. It was bisected to merge request
> commit '3424d8c18a7d ("Merge tag  'core-entry-2026-08-17' of 
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")'
> 
> Upon checking the merge, when syscall_enter_from_user_mode_randomize_stack
> fails, which could happen when a tracer like seccomp or ptrace intercepts
> and skips the syscall, the code returns to userspace immediately without
> clearing the intermediate flag which was set.
> 
> When the next syscall is made, it immediately aborts the valid syscall
> since the flag is still set. Hence clear the flag on occurrence of first
> failure.
> 
> Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
> Closes: https://lore.kernel.org/all/e301014d-568f-4ed5-bc64-b8a85ca0b1e1@linux.ibm.com/
> Fixes: 3424d8c18a7d ("Merge tag 'core-entry-2026-08-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip")
> Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
> ---

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>

Regards,
Venkat.
> PS: I have kept the block below since earlier code was checking it
> regardless of result of syscall_enter_from_user_mode. If it turns out
> to be a redundant, it can be removed later.
> 
> arch/powerpc/kernel/syscall.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
> index 4916c205c4bb..fbefe1927b10 100644
> --- a/arch/powerpc/kernel/syscall.c
> +++ b/arch/powerpc/kernel/syscall.c
> @@ -18,8 +18,10 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
> long ret;
> syscall_fn f;
> 
> - if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
> + if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
> + clear_thread_flag(TIF_SYSCALL_RET);
> return syscall_get_error(current, regs);
> + }
> 
> if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
> return syscall_get_error(current, regs);
> -- 
> 2.47.3
>
diff mbox series

Patch

diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
index 4916c205c4bb..fbefe1927b10 100644
--- a/arch/powerpc/kernel/syscall.c
+++ b/arch/powerpc/kernel/syscall.c
@@ -18,8 +18,10 @@  notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
 	long ret;
 	syscall_fn f;
 
-	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0)))
+	if (unlikely(!syscall_enter_from_user_mode_randomize_stack(regs, &r0))) {
+		clear_thread_flag(TIF_SYSCALL_RET);
 		return syscall_get_error(current, regs);
+	}
 
 	if (unlikely(test_and_clear_thread_flag(TIF_SYSCALL_RET)))
 		return syscall_get_error(current, regs);