Re: Fwd: [EXTERNAL] Re: ioctl.c:undefined reference to `__get_user_bad'

[Randy Dunlap <rdunlap@infradead.org>]

On 5/17/21 10:13 AM, Steve French wrote:
>> If you fix the issue, kindly add following tag as appropriate
>> Reported-by: kernel test robot <lkp@intel.com>
>>
>> All errors (new ones prefixed by >>):
>>
>>    arm-linux-gnueabi-ld: fs/cifs/ioctl.o: in function `cifs_dump_full_key':
>>>> ioctl.c:(.text+0x44): undefined reference to `__get_user_bad'
>>
> 
> <snip>
> 
>> # CONFIG_MMU is not set
>>
>> and arch/arm/include/asm/uaccess.h does not implement get_user(size 8 bytes)
>> for the non-MMU case:
> 
> I see another place in fs/cifs/ioctl.c where we already had been doing
> a get_user() into a u64 - any idea what you are supposed to do
> instead?  Any example code where people have worked around this.

Hi Steve,

This change in cifs_dump_full_key() makes it build OK:

-       if (get_user(suid, (__u64 __user *)arg))
+       if (get_user(suid, (unsigned int __user *)arg))


That is what the other call uses:

		case FS_IOC_SETFLAGS:
			cifs_dbg(VFS, "cifs ioctl FS_IOC_SETFLAGS:\n");
			if (pSMBFile == NULL)
				break;
			tcon = tlink_tcon(pSMBFile->tlink);
			caps = le64_to_cpu(tcon->fsUnixInfo.Capability);

			if (get_user(ExtAttrBits, (int __user *)arg)) {
				rc = -EFAULT;
				break;
			}

However, my reading/understanding is that the one immediately above
is incorrect, as is the -/+ patch above it, since get_user() gets its
data size (1, 2, 4, 8) from the type of the pointer that is passed to it.
For 8 bytes (64 bits), 'int' is not sufficient, so IMO the get_user()
call that builds:
			if (get_user(ExtAttrBits, (int __user *)arg)) {
is a bug. It should be:
			if (get_user(ExtAttrBits, (__u64 __user *)arg)) {
and if I make that latter change in the source file, the build says:

arm-linux-gnueabi-ld: fs/cifs/ioctl.o: in function `cifs_dump_full_key':
ioctl.c:(.text+0x14): undefined reference to `__get_user_bad'
arm-linux-gnueabi-ld: fs/cifs/ioctl.o: in function `cifs_ioctl':
ioctl.c:(.text+0x1f2): undefined reference to `__get_user_bad'

so now both of them fail on the get_user() of 8 bytes.

Hope that clarifies things.  It tells me that arm no-MMU still
needs support for get_user() of size 8 bytes.

-- 
~Randy