You are falling into a very common trap: plagiarism and paraphrasing are almost unrelated concepts.
Plagiarism is all about not giving credit where credit is due. This means one still may commit it even if all of the original words were changed - as long as the amount of intellectual work done by someone else going into your own work is unclear. It is mostly about ideas, but the expression (wording, sentence structure) is also important. Changing the expression is not the main reason we paraphrase, however - one could just keep the original text in quotation marks, after all, and the result would be clear of plagiarism.
Paraphrasing is there to give old ideas a new spin, to provide your own thoughts on the matter, to synthesize something new and not just give the reader a compilation of existing texts.
For your work to have substantial originality, it has to have a lot of material written by you. This means no excessive quotations (definitions of excessive vary by field). Overall, quotes should be used thoughtfully and only on concise, well-expressed thoughts. A more detailed explanation could be found here. Phrases such as "As Mark Twain notes, one does not have to rely on their memory if they are being consistently truthful" are bad writing. Paragraph-long quotes are probably unnecessary as well.
In your case, you will probably be dealing with summarizing the source material (here is another question on SE dealing with borrowing long chunks and paraphrasing).
In humanities, dealing with scarce sources is common and understandable, in STEM, it is quite rare. "A commentary on (a proof of a theorem)" is almost unheard of, but "A commentary on (a philosophy book or two)" is something ubiquitous.
If you make the amount of borrowing clear, there is no plagiarism. The criteria for originality will be field-dependent.
EDIT 2: To give a few examples of covering the existing body of material...
- As Jones and Miles explain, "Understanding the topology of contact Riemannian manifolds is essential for underwater basked weaving. Early research in underwater basked weaving has employed a number of more naïve approaches to the structural integrity such as those based on graph cuts (Xu et al., 1959), convex optimization (Nakamura, 1978) and, more recently, finite element methods (Capablanca et al., 2002; Jones et al., 2010)". This adds no value to the work of Jones and Miles, it should generally be replaced by a reference. You might also be infringing on copyright here.
- Understanding the topology of contact Riemannian manifolds is essential for underwater basked weaving (Jones and Miles, 2021). Early research in underwater basked weaving has employed a number of more naïve approaches to the structural integrity such as those based on graph cuts (Xu et al., 1959), convex optimization (Nakamura, 1978) and, more recently, finite element methods (Capablanca et al., 2002; Jones et al., 2010). And this, given the above, is outright plagiarism.
- Topology of contact Riemannian manifolds is considered indispensable for modern-day underwater basket weaving (Jones and Miles, 2021). Earlier approaches to making the baskets hold together included graph cuts (Xu et al., 1959) and finite element methods (Capablanca et al., 2002). Still plagiarism, paraphrasing probably makes it even worse.
- Consider the topology of a contact Riemannian manifold such as (...). No explanation given why it is even relevant to your research.
Jones and Miles (Jones and Miles, 2021) state: "Understanding the topology of contact Riemannian manifolds is essential for underwater basked weaving". This notion exposes the underlying trend in recent research towards making the woven baskets being more efficient in industrial applications by cutting the waste during production, increasing the surface area, and reducing the tension in the handle commonly created by machines such as WeavoTron-3000 (UBW Inc., USA). The citation is probably unneeded here, but you provide your own thoughts on the matter. If you are, say, writing a master's thesis, WeavoTron-3000 is all the rage, but there are issues which your advisor has tasked you to solve - this might be a good approach.
Underwater basket weaving as a discipline has undergone drastic changes in the past decades. Here, we follow (Jones and Miles, 2021) to track its overall history. The main issue plaguing early research on this topic was the optimal density of the twigs, which kept being inconsistent until breakthrough works of the late 1950s (Xu et al., 1959)(Stone et al., 1960). These works used graph cuts as their main approach, however, the results achieved by this technique were also highly volatile with respect to the amount of raw material used. It was solved by applying convex optimization in the now-classic paper by Nakamura (Nakamura, 1978). Industrial development in the following years was generally
highly successful, as evidenced by the growth in production capabilities (World Bank data, 2022), and new questions about the optimization arose. At the turn of the century, two main areas were considered the most prospective for future research (Ivanov et al., 2000): finite element modeling for increasing the structural stability of the baskets and contact Riemannian topology for increasing the underwater basket performance in real-world applications. First is covered by earlier and ongoing research (see e.g. Capablanca et al., 2002; Jones et al., 2010), and we are focusing on the second in this work.
As you can see, the "good" option involves a substantial expansion on whatever was expressed in any single work you have found. There is no point in following the textbook descriptions.
For theses, it is a good practice to punctuate the review section with references to increasing levels of concept difficulty: 2+2=4. The integral of a differential form over the boundary of an orientable manifold is equal to the integral of its exterior derivative over the manifold (generalized Stokes theorem; see e.g. (Tu, 2010)). A Sasakian manifold can be considered as an odd-dimensional analogue of a Kihleriaa manifold (Tanno, 1968). That way if the reader find themselves too deep into the weeds, they would have a good starting point. In that, it is similar to answering here on SE: the reader should be able to follow your text without constantly diving into the references (here it is additionally motivated by the possibility of links dying), but if a concept was already covered well elsewhere, no more than a short annotation is needed.