$D^m\cup_h D^m$, joining $D^m \amalg D^m$ along the boundary $\partial D^m$
Solution 1:
If your sphere $\Sigma(h)$ is diffeomorphic to a standard sphere, consider a diffeomorphism $\Sigma(h) \to S^m$. The two discs in $\Sigma(h)$ are smooth discs, so after applying the diffeomorphism, they're smooth discs in $S^m$. But smooth discs are tubular neighbourhoods of their centres. So they're unique up to embedded isotopy. In particular, this means that by the isotopy extension theorem you can isotope your diffeomorphism $\Sigma(h) \to S^m$ so that it sends the `bottom' disc $D^m$ in $\Sigma(h)$ to the lower hemi-sphere of $S^m$, moreover, you can ensure your diffeo $\Sigma(h) \to S^m$ is a standard diffeomorphism between the lower $D^m$ and the lower hemi-sphere. But now the upper $D^m$ in $\Sigma(h)$ is identified (via a diffeomorphism) with the upper hemi-sphere in $S^m$. So compose that map with a standard diffeomorphism between the upper-hemisphere and a $D^m$. Provided you choose it appropriately on the boundary, this is by design your extension of $h : \partial D^m \to \partial D^m$ to a diffeomorphism $\overline{h} : D^m \to D^m$.
So above, when I talk about `standard' diffeomorphisms between $D^m$ and the lower/upper hemi-spheres of $S^m$ what I mean is that $\partial D^m \times \{0\} = \partial H$ (set equality) where $H \subset S^m$ is either the upper or lower hemi-sphere in $S^m$. So to be standard I mean the diffeo must be the identity on the boundary in this sense.
Cerf went further than this, his pseudo-isotopy theorem now says that $h$ is isotopic to the identity on $\partial D^m$, provided $m \geq 6$.
Solution 2:
Topologically $\mathbb{D}^m \amalg \mathbb{D}^m / \sim$ is the m-sphere. Let's see if we can get this identification to agree with the smooth structures. Kosinski's book helped.
If $h:\partial\mathbb{D}^m \to \partial \mathbb{D}^m$ extends to a diffeomorphism $h:\mathbb{D}^m \to \mathbb{D}^m$, then there is a diffeomorphism
$$ \mathbb{D}^m \amalg_h \mathbb{D}^m \simeq \mathbb{D}^m \amalg_{id} \mathbb{D}^m \simeq S^m $$
fixing the first disk and letting $h$ act on the second disc and its boundary. This map is a homeomorphism and differentiable.
If $\mathbb{D}^m \amalg_h \mathbb{D}^m \simeq S^m $ how do we get an extension? Using whatever diffeomorphism we have, $$S^m \backslash h(\partial \mathbb{D}^m) \simeq S^m \backslash \partial \mathbb{D}^m = \mathbb{D}^m \amalg \mathbb{D}^m$$ so we get a disjoint union of two disks.
Every diffeomorphism of $S^1 \to S^1$ can be extended to the disk. Such a map as a Fourier expansion: $$ \sum a_n e^{i n \theta} \to \sum a_n r^n e^{i n \theta} \text{ with }|r|<1$$ so we have a map from $\mathbb{D} \to \mathbb{D}$ (by Cauchy-Schwarz).
Diffeomorphisms from $S^2 \to S^2$ extend to $\mathbb{D}^3$ this was proven by Munkres and by Smale.
In the comments of Tim Gower's blog, Greg Kuperberg explains that not all maps $S^6 \to S^6$ extend to $\mathbb{D}^7$. This has to do with the exotic structures on the 7-sphere, by Milnor.