On Sat, Nov 23, 2013 at 8:51 AM, Kaspar Brand <httpd-dev.2...@velox.ch>wrote:
> On 23.11.2013 13:56, Jeff Trawick wrote: > > Maybe I'm still missing something here, but it looks like we can still > > survive some calls to ssl_die() this logic. (I guess it doesn't make > sense > > to have a configuration where that could happen, but it is confusing > > looking at the code.) Look at the various calls to ap_die() in > > ssl_server_import_key(). If that happens on the SSL_AIDX_RSA call, we'll > > call it again for SSL_AIDX_DSA. I think it is best not to continue > after a > > fatal error. Also, if there is a certain type of key file and we fail > > importing it, we'll first get the fatal error message then print the > > AP01910 message then get another fatal error message. > > Ah right, I overlooked that ssl_die()s are spread all over > ssl_server_import_{cert,key}. So one option would be to make > ssl_server_import_{cert,key} sort of tri-state, and then explicitly > check for APR_EGENERAL in ssl_init_server_certs and abort in this case, > like in the attached patch? > > Kaspar > IMO it adds some future-proofing and self explanation to return early if "(rv != APR_SUCCESS && rv != APR_NOTFOUND)" instead of just checking for equality with APR_EGENERAL. YMMV :) -- Born in Roswell... married an alien... http://emptyhammock.com/