An SSL identity is characterized by four parts:
- A private key, which is not shared with anyone.
A public key, which you can share with anyone.
The private and public key form a matched pair: anything you encrypt with one can be decrypted with the other, but you cannot decrypt something encrypted with the public key without the private key or vice versa. This is genuine mathematical magic.
- Metadata attached to the public key that state who it is talking about. For a server key, this would identify the DNS name of the service that is being secured (among other things). Other data in here includes things like the intended uses (mainly used to limit the amount of damage that someone with a stolen certificate can do) and an expiry date (to limit how long a stolen certificate can be used for).
A digital signature on the combination of public key and metadata so that they can’t be messed around with and so that someone else can know how much to trust the metadata. There are multiple ways to handle who does the signature:
- Signing with the private key (from part 1, above); a self-signed certificate. Anyone can do this but it doesn’t convey much trust (precisely because anyone can do this).
- Getting a group of people who trust each other to vouch for you by signing the certificate; a web-of-trust (so called because the trust relationship is transitive and often symmetric as people sign each others certificates).
- Getting a trusted third party to do the signing; a certificate authority (CA). The identity of the CA is guaranteed by another higher-level CA in a trust chain back to some root authority that “everyone” trusts (i.e., there’s a list built into your SSL library, which it’s possible to update at deployment time).
There’s no basic technical difference between the three types of authorities above, but the nature of the trust people put in them is extremely variable. The details of why this is would require a very long answer indeed!
Items 2–4 are what comprises the digital certificate.
When the client, B, starts the SSL protocol with the server, A, the server’s digital certificate is communicated to B as part of the protocol. A’s private key is not sent, but because B can successfully decrypt a message sent by the other end with the public key in the digital certificate, B can know that A has the private key that matches. B can then look at the metadata in the certificate and see that the other end claims to be A, and can examine the signature to see how much to trust that assertion; if the metadata is signed by an authority that B trusts (directly or indirectly) then B can trust that the other end has A’s SSL identity. If that identity is the one that they were expecting (i.e., they wanted to talk to A: in practice, this is done by comparing the DNS name in the certificate with the name that they used when looking up the server address) then they can know that they have a secured communications channel: they’re good to go.
B can’t impersonate A with that information though: B doesn’t get A’s private key and so it would all fall apart at the first stage of verification. In order for some third party to impersonate B, they need to have (at least) two of:
- The private key. The owner of the identity needs to take care to stop this from happening, but it is ultimately in their hands.
- A trusted authority that makes false statements. There’s occasional weaknesses here — a self-signed authority is never very trustworthy, a web of trust runs into problems because trust is an awkward thing to handle transitively, and some CAs are thoroughly unscrupulous and others too inclined to not exclude the scum — but mostly this works fairly well because most parties are keen to not cause problems, often for financial reasons.
- A way to poison DNS so that the target believes a different server is really the one being impersonated. Without DNSsec this is somewhat easy unfortunately, but this particular problem is being reduced.
As to your other questions…
Why use hashing on the certificate if a part of it is already encrypted by the CA? Doesn’t this mean that no one can mess around with a digital certificate (in high probability) anyway?
While keys are fairly long, certificates are longer (for one thing, they include the signers public key anyway, which is typically the same length the key being signed). Hashing is part of the general algorithm for signing documents anyway because nobody wants to be restricted to signing only very short things. Given that the algorithm is required, it makes sense to use it for this purpose.
If I am stackoverflow and I have 3 servers doing the same thing – allowing clients to access, read, identify etc. – do I have to have a different digital certificate for each of the 3 servers.
If you have several servers serving the same DNS name (there’s many ways to do this, one of the simplest being round-robin DNS serving) you can put the same identity on each of them. This slightly reduces security, but only very slightly; it’s still one service that just happens to be implemented by multiple servers. In theory you could give each one a different identity (though with the same name) but I can’t think of any good reason for actually doing it; it’s more likely to worry people than the alternative.
Also note that it’s possible to have a certificate for more than one service name at once. There are two mechanisms for doing this (adding alternate names to the certificate or using a wildcard in the name in the certificate) but CAs tend to charge quite a lot for signing certificates with them in.