linux-kernel - Re: [PATCH 2/2] riscv: uaccess: do not do misaligned accesses in get/put

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 for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <cea8cdf2-4b1e-4628-a5df-249adf4387b5@rivosinc.com>
Date: Mon, 2 Jun 2025 09:37:08 +0200
From: Clément Léger <cleger@...osinc.com>
To: Alexandre Ghiti <alex@...ti.fr>, linux-riscv@...ts.infradead.org,
 linux-kernel@...r.kernel.org
Cc: Paul Walmsley <paul.walmsley@...ive.com>,
 Palmer Dabbelt <palmer@...belt.com>, Albert Ou <aou@...s.berkeley.edu>
Subject: Re: [PATCH 2/2] riscv: uaccess: do not do misaligned accesses in
 get/put_user()



On 31/05/2025 14:35, Alexandre Ghiti wrote:
> On 5/30/25 22:56, Clément Léger wrote:
>> Doing misaligned access to userspace memory would make a trap on
>> platform where it is emulated. Latest fixes removed the kernel
>> capability to do unaligned accesses to userspace memory safely since
>> interrupts are kept disabled at all time during that. Thus doing so
>> would crash the kernel.
>>
>> Such behavior was detected with GET_UNALIGN_CTL() that was doing
>> a put_user() with an unsigned long* address that should have been an
>> unsigned int*. Reenabling kernel misaligned access emulation is a bit
>> risky and it would also degrade performances. Rather than doing that,
>> we will try to avoid any misaligned accessed by using copy_from/to_user()
>> which does not do any misaligned accesses. This can be done only for
>> !CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS and thus allows to only generate
>> a bit more code for this config.
>>
>> Signed-off-by: Clément Léger <cleger@...osinc.com>
>> ---
>>   arch/riscv/include/asm/uaccess.h | 28 ++++++++++++++++++++++------
>>   1 file changed, 22 insertions(+), 6 deletions(-)
>>
>> diff --git a/arch/riscv/include/asm/uaccess.h b/arch/riscv/include/
>> asm/uaccess.h
>> index 046de7ced09c..b542c05f394f 100644
>> --- a/arch/riscv/include/asm/uaccess.h
>> +++ b/arch/riscv/include/asm/uaccess.h
>> @@ -169,8 +169,21 @@ do {                                \
>>     #endif /* CONFIG_64BIT */
>>   +unsigned long __must_check __asm_copy_to_user(void __user *to,
>> +    const void *from, unsigned long n);
>> +unsigned long __must_check __asm_copy_from_user(void *to,
>> +    const void __user *from, unsigned long n);
>> +
>>   #define __get_user_nocheck(x, __gu_ptr, label)            \
>>   do {                                \
>> +    if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
>> {            \
>> +        if (!IS_ALIGNED((uintptr_t)__gu_ptr, sizeof(*__gu_ptr)))
>> {        \
> 
> 
> Nit: I would use && instead of 2 ifs.
> 
> 
>> +            if (__asm_copy_from_user(&(x), __gu_ptr,
>> sizeof(*__gu_ptr)))    \
>> +                goto label;            \
>> +            else                    \
>> +                break;                \
> 
> 
> Here I would remove the else

Hi Alex,

The "else" is needed to break from the outer do/while loop or it will go
though the next switch case (and it will crash due to misaligned accesses).

> 
> 
>> +        }                        \
>> +    }                            \
>>       switch (sizeof(*__gu_ptr)) {                \
>>       case 1:                            \
>>           __get_user_asm("lb", (x), __gu_ptr, label);    \
>> @@ -297,6 +310,15 @@ do {                                \
>>     #define __put_user_nocheck(x, __gu_ptr, label)            \
>>   do {                                \
>> +    if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
>> {            \
>> +        if (!IS_ALIGNED((uintptr_t)__gu_ptr, sizeof(*__gu_ptr)))
>> {        \
>> +            unsigned long val = (unsigned long)(x);                \
> 
> 
> Here it sems like __inttype(*(__gu_ptr)) is more accurate than unsigned
> long, even though I think unsigned long works fine too.

Wasn't aware of __inttype, but it sounds good.

Thanks,

Clément

> 
> 
>> +            if (__asm_copy_to_user(__gu_ptr, &(val),
>> sizeof(*__gu_ptr)))    \
>> +                goto label;            \
>> +            else                    \
>> +                break;                \
>> +        }                        \
>> +    }                            \
>>       switch (sizeof(*__gu_ptr)) {                \
>>       case 1:                            \
>>           __put_user_asm("sb", (x), __gu_ptr, label);    \
>> @@ -385,12 +407,6 @@ err_label:                            \
>>           -EFAULT;                    \
>>   })
>>   -
>> -unsigned long __must_check __asm_copy_to_user(void __user *to,
>> -    const void *from, unsigned long n);
>> -unsigned long __must_check __asm_copy_from_user(void *to,
>> -    const void __user *from, unsigned long n);
>> -
>>   static inline unsigned long
>>   raw_copy_from_user(void *to, const void __user *from, unsigned long n)
>>   {