[PATCH v2 01/50] lib: Add memdup()

[Heinrich Schuchardt]

On 5/10/21 11:00 AM, Rasmus Villemoes wrote:
> On 06/05/2021 19.41, Simon Glass wrote:
>> Hi Pratyush,
>>
>> On Thu, 6 May 2021 at 10:07, Pratyush Yadav <p.yadav at ti.com> wrote:
>>>
>>> On 06/05/21 08:23AM, Simon Glass wrote:
>>>> Add a function to duplicate a memory region, a little like strdup().
>>>>
>>>> Signed-off-by: Simon Glass <sjg at chromium.org>
>>>> ---
>>>>
>>>> Changes in v2:
>>>> - Add a patch to introduce a memdup() function
>>>>
>>>>   include/linux/string.h | 13 +++++++++++++
>>>>   lib/string.c           | 13 +++++++++++++
>>>>   test/lib/string.c      | 32 ++++++++++++++++++++++++++++++++
>>>>   3 files changed, 58 insertions(+)
>>>>
>>>> diff --git a/include/linux/string.h b/include/linux/string.h
>>>> index dd255f21633..3169c93796e 100644
>>>> --- a/include/linux/string.h
>>>> +++ b/include/linux/string.h
>>>> @@ -129,6 +129,19 @@ extern void * memchr(const void *,int,__kernel_size_t);
>>>>   void *memchr_inv(const void *, int, size_t);
>>>>   #endif
>>>>
>>>> +/**
>>>> + * memdup() - allocate a buffer and copy in the contents
>>>> + *
>>>> + * Note that this returns a valid pointer even if @len is 0
>>>
>>> I'm uneducated about U-Boot's memory allocator. But I wonder how it
>>> returns a valid pointer even on 0 length allocations. What location does
>>> it point to? What are users expected to do with that pointer? They
>>> obviously can't read/write to it since it is supposed to be a 0 byte
>>> long allocation. If another positive length allocation happens before
>>> the said pointer is freed, will it point to the same memory location? If
>>> not, isn't the 0-length pointer actually at least a 1-length pointer?
>>
>> I think it is just a 0-length pointer and that the only thing you can
>> do with it is call free().
>>
>> I am certainly no expert on this sort of thing though. It seems that
>> some implementations return NULL for a zero size, some return a valid
>> pointer which can be passed to free().
>
> It's implementation-defined, which means that one cannot know for
> certain that a given malloc implementation won't return NULL for a
> request of 0 bytes. The linux kernel solved that problem by introducing
> ZERO_SIZE_PTR which is basically just (void*)16L or something like that
> - that way callers don't have to write their "did the allocation
> succeed" test in the ugly
>
>    if (!p && size != 0)
>      error_out;
>
> way. Of course kfree() must then accept that in addition to NULL, but
> it's not really more expensive to have that early nop check be
>
>    if ((unsigned long)ptr <= 16)
>       return;
>
> instead of
>
>    if (!ptr)
>      return;
>
>
> "man malloc" says
>
> RETURN VALUE
>         The malloc() and calloc() functions return a pointer to the
> allocated memory, which is suitably aligned for any built-in type.  On
> error,  these  functions
>         return  NULL.   NULL may also be returned by a successful call to
> malloc() with a size of zero, or by a successful call to calloc() with
> nmemb or size equal
>         to zero.
>
>
> Anyway, I don't think this helper should be put in string.c - it needs
> to be in some C file that's easily compiled for both board, sandbox and
> host tools (for the latter probably via the "tiny one-line wrapper that
> just includes the whole real C file"). I see there's linux_string.c already.
>
> Rasmus
>

Our malloc() implementation allocates space for metadata of type struct
malloc_chunk even if the argument is zero.

This metadata is used to check that a call to free() refers to a valid
pointer.

I don't see a need to change this behavior.

Best regards

Heinrich