We assume you have a secure hash function $H: \{0,1\}^{128} \rightarrow\{0,1\}^{128}$. Using this you want to construct a different hash function $H' = H(m \oplus s)$, where $m$ is your plaintext and $s$ is the salt.
As @frigeu already mentioned, it may be collision-resistant and preimage resistant. If an adversary knows $s$, it is not a secure PRF. If $s$ is not know, you can imagine it as a keyed Hash function, which can result in a PRF.
But I want to go into detail with salt and preimage resistance. In computer science salt is used to describe a technique to store passwords in a more secure way.
Imagine a data base, where you store users and their passwords. In general it is a very bad idea to store them directly. In order to solve this, the user input is hashed and the hashed values are stored.
But now you can assume an adversary obtaining the data base. The adversary can precompute all common password (e.g. 12345, password) and simply check for similarities. A more advanced adversary can use rainbow tables. The result is, that a lot of passwords will be broken.
With a salt you try to avoid such attacks, by storing the salt values next to the user data, and then use $H'$ to compute the hashed password you want to store. This leads to a harder-to-attack problem for the adversary, even if the salt is known, because he has to recompute the rainbow tables and can not use the same. (Note: You can increase the security again, using pepper. Here you can assume, that the value is not obtained by the adversary.)
So the question is not, weather you are secure or not. The question is, against which adversary and adversary goals you archive security.
Or simply put: How difficult do you make it for the adversary?