crypto: build warning fixes

[brody4hire]

Checklist

• make -j8 test (UNIX [macOS]; Linux i386 & amd64) and vcbuild test nosign (.\vcbuild.bat nosign then .\vcbuild.bat test nosign nobuild as Administrator on Windows) PASS OK
• commit message follows commit guidelines

Additional item

• I hereby certify the statements in Developer's Certificate of Origin 1.1

Affected core subsystem(s)

crypto

Description of change

• use unsigned long exponent in BIO_printf: resolve warning in Ubuntu Linux i386 build
• use static kTLSTicketSizeMask: resolve conversion warning on Windows

NOTE: This was originally a part of PR #10139 (build, warning, header, and include fixes).

ADDITIONAL NOTE: This and other changes related to PR #10139 are marked // TBD POSSIBLE DATA LOSS: since I think we should examine these conversions at some point. I would be happy to take some or all of these out if necessary.

[bnoordhuis]

This is not correct per se on 32 bits architectures where long is 32 bits and BN_ULONG is a 64 bits unsigned long long.

[brody4hire]

I will fix this tomorrow.

[bnoordhuis]

@indutny Is there a reason tls_ticket_size_ is of type uint16_t instead of size_t?

[indutny]

That's how much bits it occupies in binary representation. I don't have any strong feelings about making it wider, though.

[brody4hire]

Here is what I gather from a quick grep: tls_ticket_size_ stores a ticket size and tls_ticket_ stores a pointer to some ticket information but they are only used to determine a hello.has_ticket_ boolean value. I start to wonder if we could replace the tls_ticket_ and tls_ticket_size_ members with a member like has_valid_tls_ticker_ to avoid keeping more state than necessary?

[bnoordhuis]

@indutny?

[brody4hire]

In src/node_crypto.c > @@ -1536,8 +1536,9 @@ static Local<Object> X509ToObject(Environment*

env, X509* cert) { String::kNormalString, mem->length)); (void) BIO_reset(bio); - BN_ULONG exponent_word = BN_get_word(rsa->e); - BIO_printf(bio, "0x%lx", exponent_word); + unsigned long exponent = // NOLINT(runtime/int) + static_cast<unsigned long>(BN_get_word(rsa->e)); // NOLINT(runtime/int)

This is not correct per se on 32 bits architectures where long is 32 bits

and BN_ULONG is a 64 bits unsigned long long. I will look at this.

In src/node_crypto_clienthello.cc: > @@ -146,7 +149,8 @@ void ClientHelloParser::ParseExtension(const

uint16_t type, ocsp_request_ = 1; break; case kTLSSessionTicket: - tls_ticket_size_ = len; + // TBD POSSIBLE DATA LOSS: + tls_ticket_size_ = static_cast<uint16_t>(len & kTLSTicketSizeMask);

@indutny Is there a reason tls_ticket_size_ is of type uint16_t instead

of size_t? :+1: for fixing on my part. @bnoorduis I will look at these and your comments on the other issues when I get home tonight or maybe tomorrow morning. Thanks for the feedback!

[brody4hire]

Closing for now, I would like to deal with this later.