Hello Liu,
On Fri, 2020-07-31 at 18:41 +0800, Zhiqiang Liu wrote:
> In vector_alloc_slot func, if REALLOC fails, it means new slot
> allocation fails. However, it just update v->allocated and then
> return the old v->slot without new slot. So, the caller will take
> the last old slot as the new allocated slot, and use it by calling
> vector_set_slot func. Finally, the data of last slot is lost.
>
> Here, if REALLOC or MALLOC fails, we will return NULL.
>
> Signed-off-by: Zhiqiang Liu <liuzhiqiang26@xxxxxxxxxx>
> Signed-off-by: lixiaokeng <lixiaokeng@xxxxxxxxxx>
> ---
> libmultipath/vector.c | 10 ++++++----
> 1 file changed, 6 insertions(+), 4 deletions(-)
>
> diff --git a/libmultipath/vector.c b/libmultipath/vector.c
> index 501cf4c5..29dc9848 100644
> --- a/libmultipath/vector.c
> +++ b/libmultipath/vector.c
> @@ -49,12 +49,14 @@ vector_alloc_slot(vector v)
> else
> new_slot = (void *) MALLOC(sizeof (void *) * v-
> >allocated);
>
> - if (!new_slot)
> + /* If REALLOC or MALLOC fails, it means new slot allocation
> fails, so return NULL. */
> + if (!new_slot) {
> v->allocated -= VECTOR_DEFAULT_SIZE;
> - else
> - v->slot = new_slot;
> + return NULL;
> + }
>
> - return v->slot;
> + v->slot = new_slot;
> + return v->slot[VECTOR_SIZE(v) - 1];
This changes the semantics of the function by returning the last
element of the vector rather than v->slot. That's dangerous because
these elements aren't initialized. You might as well return NULL in
case of success, which would obviously be wrong (actually, new elements
_should_ be initialized to NULL). As the return value is only ever used
to check for successful allocation, it might be best to change it into
a bool, avoiding any ambiguity about its meaning.
If you want to clean up this function (appreciated!), please do it
right:
- increment v->allocated only after successful allocation,
- avoid the "if (v->slot)" conditional by just calling realloc(),
- make sure all newly allocated vector elements are set to NULL,
- optionally, change return value to bool (see above).
Regards,
Martin
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
