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@efficios.com>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.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);
+
 	/*
 	 * Additional feature fields added after ORIG_RSEQ_SIZE
 	 * need to be conditionally updated only if
 	 * t->rseq_len != ORIG_RSEQ_SIZE.
 	 */
 	user_write_access_end();
-	rseq_set_ro_fields(t, cpu_id, cpu_id, node_id, mm_cid);
 	trace_rseq_update(t);
 	return 0;
 
@@ -195,6 +201,7 @@ static int rseq_update_cpu_node_id(struct task_struct *t)
 
 static int rseq_reset_rseq_cpu_node_id(struct task_struct *t)
 {
+	struct rseq __user *rseq = t->rseq;
 	u32 cpu_id_start = 0, cpu_id = RSEQ_CPU_ID_UNINITIALIZED, node_id = 0,
 	    mm_cid = 0;
 
@@ -202,38 +209,36 @@ static int rseq_reset_rseq_cpu_node_id(struct task_struct *t)
 	 * Validate read-only rseq fields.
 	 */
 	if (rseq_validate_ro_fields(t))
-		return -EFAULT;
-	/*
-	 * Reset cpu_id_start to its initial state (0).
-	 */
-	if (put_user(cpu_id_start, &t->rseq->cpu_id_start))
-		return -EFAULT;
-	/*
-	 * Reset cpu_id to RSEQ_CPU_ID_UNINITIALIZED, so any user coming
-	 * in after unregistration can figure out that rseq needs to be
-	 * registered again.
-	 */
-	if (put_user(cpu_id, &t->rseq->cpu_id))
-		return -EFAULT;
-	/*
-	 * Reset node_id to its initial state (0).
-	 */
-	if (put_user(node_id, &t->rseq->node_id))
-		return -EFAULT;
+		goto efault;
+
+	if (!user_write_access_begin(rseq, t->rseq_len))
+		goto efault;
+
 	/*
-	 * Reset mm_cid to its initial state (0).
+	 * Reset all fields to their initial state.
+	 *
+	 * All fields have an initial state of 0 except cpu_id which is set to
+	 * RSEQ_CPU_ID_UNINITIALIZED, so that any user coming in after
+	 * unregistration can figure out that rseq needs to be registered
+	 * again.
 	 */
-	if (put_user(mm_cid, &t->rseq->mm_cid))
-		return -EFAULT;
-
-	rseq_set_ro_fields(t, cpu_id_start, cpu_id, node_id, mm_cid);
+	unsafe_rseq_set_field(t, cpu_id_start, cpu_id_start, 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);
 
 	/*
 	 * Additional feature fields added after ORIG_RSEQ_SIZE
 	 * need to be conditionally reset only if
 	 * t->rseq_len != ORIG_RSEQ_SIZE.
 	 */
+	user_write_access_end();
 	return 0;
+
+efault_end:
+	user_write_access_end();
+efault:
+	return -EFAULT;
 }
 
 static int rseq_get_rseq_cs(struct task_struct *t, struct rseq_cs *rseq_cs)
-- 
2.43.0

* Michael Jeanson <mjeanson@efficios.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@efficios.com>
> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.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()?

Thanks,

	Ingo

On 2025-02-22 08:27, Ingo Molnar wrote:
> 
> * Michael Jeanson <mjeanson@efficios.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@efficios.com>
>> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.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

* Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:

> > > -	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.

Absolutely.

> 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.

Yeah, put_user()'s oddity is a random historic idiosyncrasy that we 
want to preserve in derived naming to reduce the potential for 
confusion.

Thanks,

	Ingo

On 2025-02-22 09:12, Ingo Molnar wrote:
> 
> * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> 
>>>> -	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.
> 
> Absolutely.
> 
>> 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.
> 
> Yeah, put_user()'s oddity is a random historic idiosyncrasy that we 
> want to preserve in derived naming to reduce the potential for 
> confusion.

Would that be ok?

  rseq_unsafe_put_user(t, value, field, error_label)

> 
> Thanks,
> 
> 	Ingo

* Michael Jeanson <mjeanson@efficios.com> wrote:

> >> 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.
> > 
> > Yeah, put_user()'s oddity is a random historic idiosyncrasy that we 
> > want to preserve in derived naming to reduce the potential for 
> > confusion.
> 
> Would that be ok?
> 
>   rseq_unsafe_put_user(t, value, field, error_label)

Yeah, I think so.

Thanks,

	Ingo