In the paper of "Reaction Attacks against Several Public-Key Cryptosystems" CiteSeerX link, reaction attack is defined informally as "Obtaining information about the private key or plaintext by watching the reaction of someone decrypting a given ciphertext with the private key."
Is reaction attack explicitly defined in literature? What is the difference between fault attack and reaction attack -as defined here- ?
"Reaction attack" seems to be just a custom name used in a few papers, meaning the reaction of the decryption oracle on maliciously crafted/modified ciphertexts. These are just CCA attacks, not side-channel attacks a priori, but in some cases side channel information such as timing can be used.
These attacks are based exploiting the decryption oracle. Note that many CPA-secure schemes are not CCA-secure (e.g. CBC encryption of a block cipher is vulnerable to the padding oracle attack), however there are ways to convert them in CCA-secure schemes (e.g. adding a MAC for symmetric encryption, or the Fujisaki-Okamoto (FO) transformation for asymmetric schemes).
