Re: Proposed patch for key managment

On Thu, Dec 10, 2020 at 07:26:48PM +0800, Neil Chen wrote:
>
>
> Hi, everyone
>
> I have read the patch and did some simple tests. I'm not entirely sure
> about some code segments; e.g.:
>
> In the BootStrapKmgr() we generate a data encryption key by:
> key = generate_crypto_key(file_encryption_keylen);
>
> However, I found that the file_encryption_keylen is always 0 in bootstrap
> mode because there exitst another variable bootstrap_file_encryption_keylen
> in xlog.c and bootstrap.c.

Oh, good point; that is very helpful. I was relying on SetConfigOption
to set file_encryption_keylen, but that happens _after_ we create the
keys, so they were zero length. I have fixed this by passing
bootstrap_file_encryption_keylen to the boot routines. The diff URL has
the fix:

https://github.com/postgres/postgres/compare/master...bmomjian:key.diff

> We get the REL/WAL key by KmgrGetKey() call and it works like:
> return (const CryptoKey *) &(KmgrShmem->intlKeys[id]);
>
> But in bootstrap mode, the KmgrShmem are not assigned. So, if we want to
> use it to encrypt something in bootstrap mode, I suggest we make the
> following changes:
> if ( in bootstrap mode)
> return intlKeys[id]; // a static variable which contains key
> else
> reutrn (const CryptoKey *) &(KmgrShmem->intlKeys[id]);

Yes, you are also correct here. I had not gotten to using KmgrGetKey
yet, but it clearly needs your suggestion, so have done that.

Thanks for your help.

--
Bruce Momjian <bruce(at)momjian(dot)us> https://momjian.us
EnterpriseDB https://enterprisedb.com

The usefulness of a cup is in its emptiness, Bruce Lee