Re: [PATCH 6/6] kselftest/arm64: Check mte tagged user address in kernel

[Amit Kachhap]

On 9/22/20 4:11 PM, Catalin Marinas wrote:
> On Tue, Sep 01, 2020 at 02:57:19PM +0530, Amit Daniel Kachhap wrote:
> > Add a testcase to check that user address with valid/invalid
> > mte tag works in kernel mode. This test verifies the kernel API's
> > __arch_copy_from_user/__arch_copy_to_user works by considering
> > if the user pointer has valid/invalid allocation tags.
> >
> > In MTE sync mode a SIGSEV fault is generated if a user memory
> > with invalid tag is accessed in kernel. In async mode no such
> > fault occurs.
>
> We don't generate a SIGSEGV for faults in the uaccess routines. The
> kernel simply returns less copied bytes than what was requested or -1
> and setting errno.

ok. I will update in the next iteration.
> BTW, Qemu has a bug and it reports the wrong exception class (lower
> DABT) for a tag check fault while in the uaccess routines, leading to
> kernel panic (bad mode in synchronous abort handler).

Yes I am also seeing this.
> > +static int check_usermem_access_fault(int mem_type, int mode, int mapping)
> > +{
> > +	int fd, ret, i, err;
> > +	char val = 'A';
> > +	size_t len, read_len;
> > +	void *ptr, *ptr_next;
> > +	bool fault;
> > +
> > +	len = 2 * page_sz;
> > +	err = KSFT_FAIL;
> > +	/*
> > +	 * Accessing user memory in kernel with invalid tag should fault in sync
> > +	 * mode but may not fault in async mode as per the implemented MTE
> > +	 * support in Arm64 kernel.
> > +	 */
> > +	if (mode == MTE_ASYNC_ERR)
> > +		fault = false;
> > +	else
> > +		fault = true;
> > +	mte_switch_mode(mode, MTE_ALLOW_NON_ZERO_TAG);
> > +	fd = create_temp_file();
> > +	if (fd == -1)
> > +		return KSFT_FAIL;
> > +	for (i = 0; i < len; i++)
> > +		write(fd, &val, sizeof(val));
> > +	lseek(fd, 0, 0);
> > +	ptr = mte_allocate_memory(len, mem_type, mapping, true);
> > +	if (check_allocated_memory(ptr, len, mem_type, true) != KSFT_PASS) {
> > +		close(fd);
> > +		return KSFT_FAIL;
> > +	}
> > +	mte_initialize_current_context(mode, (uintptr_t)ptr, len);
> > +	/* Copy from file into buffer with valid tag */
> > +	read_len = read(fd, ptr, len);
> > +	ret = errno;
>
> My reading of the man page is that errno is set only if read() returns
> -1.

Yes. The checks should be optimized here.
> > +	mte_wait_after_trig();
> > +	if ((cur_mte_cxt.fault_valid == true) || ret == EFAULT || read_len < len)
> > +		goto usermem_acc_err;
> > +	/* Verify same pattern is read */
> > +	for (i = 0; i < len; i++)
> > +		if (*(char *)(ptr + i) != val)
> > +			break;
> > +	if (i < len)
> > +		goto usermem_acc_err;
> > +
> > +	/* Tag the next half of memory with different value */
> > +	ptr_next = (void *)((unsigned long)ptr + page_sz);
> > +	ptr_next = mte_insert_tags(ptr_next, page_sz);
> > +	if (!ptr_next)
> > +		goto usermem_acc_err;
> > +	lseek(fd, 0, 0);
> > +	/* Copy from file into buffer with invalid tag */
> > +	read_len = read(fd, ptr, len);
> > +	ret = errno;
> > +	mte_wait_after_trig();
> > +	if ((fault == true) &&
>
> Nitpick: just use "if (fault &&), it's a bool already.

ok.
> > +	    (cur_mte_cxt.fault_valid == true || ret == EFAULT || read_len < len)) {
> > +		err = KSFT_PASS;
> > +	} else if ((fault == false) &&
> > +		   (cur_mte_cxt.fault_valid == false && read_len == len)) {
>
> Same here, !fault, !cur_mte_cxt.fault_valid.

ok.
>