More various fixes to reports from static analyzers

[gasbytes]

• add esp encapsulation to tcp_send_empty_immediate so pure acks on ipsec
• basedd connections are not sent in plaintext when the tx fifo is full
• (and the link layer is being used)
• add esp encapsulation to tcp_send_zero_wnd_probe so zero_window probes on ipsec-protected connections are not sent in plaintext
• add esp encapsulation to wolfip_forwad_packet so forwarded ip packets
• are not sent in plaintext when the egresse interface has esp sa
• ocnfigureed
• reset sift-sdown cursor to root on each iteration of
• timers_binheapp_poop so that skipping cancelled timers does not b reak
• the min-heap invariant
• send syn-ack immediately in the listen to syn_rcvd transition instead of deferring it until accept() or the ctrl_rto timer fires.
• validate arp reply sender ip against broadcast, multicast, zero and own address before caching, matching the existing arp request handler checks
• deconfigure the offered ip address when dhcp request retries are exhausted so the device does not keep using unconfirmed lease.
• extend multicast classification tests to cover the 225-239 range and boundaries so mask constant mutations are detected
• add ssrr source routine drop test to cover 0x89 variant type branch alongside the existing lsrr test
• add multicast source address test for ip_recv to cover the wolfip_ip_is_multicast branch of the rfc 1122 source validation
• add multicast sender test for arp request validation to cover the wolfip_ip_is_multicast branch of the cache poisoning guard.
• add esp encapsulation to tcp_send_reset_reply so rst segments toward esp_protected peers are not sent in plaintext
• nclude tcp timestamp option in zero-window probes when timestamps are negotiated, per rfc 7323 (around section 3.2)
• process ack+fin segments in syn_rcvd state per rfc 9293 instead of silently discarding non-pure acks.
• Compute ICMP TTL exceeded and destination unreachable original packet
• quote size from the actual IP header length so packets with IP options
• include the full header plus 8 bytes of transport data.
• seed ipcounter with a random value at init so ip identification fields on non-df packets are not predictable from zero
• set the df bit on tcp rst replies to match the normal tcp output patch and prevent ip id observability
• zero hmac hash buffers on the stack after icv comparison and copy to prevent residual keyed hash output from persisting
• zero gcm/gmac nonce buffers on the stack after use to prevent residual implicit salt bytes from persisting

[wolfSSL-Fenrir-bot]

Fenrir Automated Review — PR #100

Scan targets checked: wolfip-bugs, wolfip-compliance, wolfip-defaults, wolfip-mutation, wolfip-proptest, wolfip-src

No new issues found in the changed files. ✅

[philljj]

I don't think these wc_ForceZero(hash, ...); are needed either. The hash ICVs are public values calculated from public payloads.

If the attacker wants samples of [msg, icv] tuples they can just passively monitor valid ESP traffic.