linux-kernel - Re: [PATCH] rseq: update kernel fields in lockstep with CONFIG_DEBUG

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <974359d5-43f3-483a-89cf-79e9b4965785@efficios.com>
Date: Sat, 22 Feb 2025 08:56:50 -0500
From: Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
To: Ingo Molnar <mingo@...nel.org>, Michael Jeanson <mjeanson@...icios.com>
Cc: Peter Zijlstra <peterz@...radead.org>,
 "Paul E. McKenney" <paulmck@...nel.org>, Boqun Feng <boqun.feng@...il.com>,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH] rseq: update kernel fields in lockstep with
 CONFIG_DEBUG_RSEQ

On 2025-02-22 08:27, Ingo Molnar wrote:
> 
> * Michael Jeanson <mjeanson@...icios.com> wrote:
> 
>> With CONFIG_DEBUG_RSEQ an in-kernel copy of the read-only fields is
>> kept synchronized with the user-space fields. Ensure the updates
>> are done in lockstep in case we error out on a write to user-space.
>>
>> Fixes: 7d5265ffcd8b ("rseq: Validate read-only fields under DEBUG_RSEQ config")
>> Signed-off-by: Michael Jeanson <mjeanson@...icios.com>
>> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
>> ---
>>   kernel/rseq.c | 85 +++++++++++++++++++++++++++------------------------
>>   1 file changed, 45 insertions(+), 40 deletions(-)
>>
>> diff --git a/kernel/rseq.c b/kernel/rseq.c
>> index 2cb16091ec0a..5bdb96944e1f 100644
>> --- a/kernel/rseq.c
>> +++ b/kernel/rseq.c
>> @@ -26,6 +26,11 @@
>>   				  RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL | \
>>   				  RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE)
>>   
>> +static struct rseq __user *rseq_user_fields(struct task_struct *t)
>> +{
>> +	return (struct rseq __user *) t->rseq;
>> +}
>> +
>>   #ifdef CONFIG_DEBUG_RSEQ
>>   static struct rseq *rseq_kernel_fields(struct task_struct *t)
>>   {
>> @@ -78,24 +83,24 @@ static int rseq_validate_ro_fields(struct task_struct *t)
>>   	return -EFAULT;
>>   }
>>   
>> -static void rseq_set_ro_fields(struct task_struct *t, u32 cpu_id_start, u32 cpu_id,
>> -			       u32 node_id, u32 mm_cid)
>> -{
>> -	rseq_kernel_fields(t)->cpu_id_start = cpu_id;
>> -	rseq_kernel_fields(t)->cpu_id = cpu_id;
>> -	rseq_kernel_fields(t)->node_id = node_id;
>> -	rseq_kernel_fields(t)->mm_cid = mm_cid;
>> -}
>> +/*
>> + * Update an rseq field and its in-kernel copy in lock-step to keep a coherent
>> + * state.
>> + */
>> +#define unsafe_rseq_set_field(t, field, value, error_label)		\
>> +	do {								\
>> +		unsafe_put_user(value, &rseq_user_fields(t)->field, error_label);	\
>> +		rseq_kernel_fields(t)->field = value;			\
>> +	} while (0)
>> +
>>   #else
>>   static int rseq_validate_ro_fields(struct task_struct *t)
>>   {
>>   	return 0;
>>   }
>>   
>> -static void rseq_set_ro_fields(struct task_struct *t, u32 cpu_id_start, u32 cpu_id,
>> -			       u32 node_id, u32 mm_cid)
>> -{
>> -}
>> +#define unsafe_rseq_set_field(t, field, value, error_label)		\
>> +	unsafe_put_user(value, &rseq_user_fields(t)->field, error_label)
>>   #endif
>>   
>>   /*
>> @@ -173,17 +178,18 @@ static int rseq_update_cpu_node_id(struct task_struct *t)
>>   	WARN_ON_ONCE((int) mm_cid < 0);
>>   	if (!user_write_access_begin(rseq, t->rseq_len))
>>   		goto efault;
>> -	unsafe_put_user(cpu_id, &rseq->cpu_id_start, efault_end);
>> -	unsafe_put_user(cpu_id, &rseq->cpu_id, efault_end);
>> -	unsafe_put_user(node_id, &rseq->node_id, efault_end);
>> -	unsafe_put_user(mm_cid, &rseq->mm_cid, efault_end);
>> +
>> +	unsafe_rseq_set_field(t, cpu_id_start, cpu_id, efault_end);
>> +	unsafe_rseq_set_field(t, cpu_id, cpu_id, efault_end);
>> +	unsafe_rseq_set_field(t, node_id, node_id, efault_end);
>> +	unsafe_rseq_set_field(t, mm_cid, mm_cid, efault_end);
> 
> Could we please name the new wrapper rseq_unsafe_put_user(), to make it
> clear it's a wrapper around unsafe_put_user()?

If we do this then need to make sure the order of arguments becomes
similar to unsafe_put_user for consistency. If we look at its macro:

#define unsafe_put_user(x,p,e) unsafe_op_wrap(__put_user(x,p),e),
we have the following argument order:

1) value (x),
2) pointer (p),
3) error label (e).

So far unsafe_rseq_set_field has:

1) task struct pointer,
2) rseq field name,
3) value,
4) error label.

I always find it odd that the "source" argument comes first and
the "destination" argument comes second in all put_user() APIs,
compared to memcpy, WRITE_ONCE() and all assignments (e.g.
operator "=" LHS vs RHS). Choosing a different argument order
therefore made sense with a naming different from "*put_user",
but not so much if we use a derived naming.

This argument order oddness was actually one motivation for using
a naming *different* from put_user.

We can consider "task" a context, "field" as destination
argument, and "value" a source argument. So if we use a
similar approach as put_user, we'd have "task" as first
argument (context), "value" as second argument, followed
by "field", then error label.

Thoughts ?

Thanks,

Mathieu

> 
> Thanks,
> 
> 	Ingo


-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com