* https-proxy-config: (92 commits)
increases tcp proxy listener timeout to 2 minutes
fixes dumb bug in tcp reverse proxy remote address resolution
upgrades libh2o to v0.13.5 (h2o v2.2.5)
sets the http client request timeout to 2 minutes
adds an http request timer to ensure connections are always closed
adds an http request timer to ensure connections are always closed
adds (disabled) h2o access log
don't leak ship-name in _proxy_ward_connect()
refactors .http.ports write/release (avoid the loom, etc.)
updates SSL_CTX init to support any PEM private key, simplifies error handling
adds u3_wcon (ward candidate conn), refactors ward connection accept and auth
enables ALPN and HTTP2 over TLS
increases timeout before server struct, adds explanatory comment
closes u3_ward stream and timer sequentially, fixing use-after-free per ASan
moves u3_pact to vere.h, replacing the unused u3_apac
fixes memory leak in failed cttp.c address resolution
suppresses AddressSanitizer in vararg functions u3i_molt and u3r_mean
ignores AddressSanitizer errors in jets.c
updates %wise and %that to give and receive the proxy auth nonce
implements nonce-baced authentication for u3_ward proxies
...
* master:
adds an http request timer to ensure connections are always closed
moves u3_pact to vere.h, replacing the unused u3_apac
fixes memory leak in failed cttp.c address resolution
Remove deprecated paths from gitignore
print error on failed syscall in _ce_patch_create()
removes tabs from http.c
removes orphaned telnet constants and forward declarations
process sampling already takes care to turn off cpu profiling during its
sampling. this "fix" for mid-equality sampling was masking another bug
at an earlier point in the debugging process for the unifying equality
changes. I realized upon reflection that it was incorrect.
Previously, trimming the penultimate value from a hash bucket would
leave arm_u in an inconsistent state: mug_w would be pointing at
a key-value pair in a node, but arm_u.buc_o and arm_u.inx_w were
not updated to reflect this. Correspondingly, a rare edge case could
occur where said key-value pair would not be marked cold (due to
buc_o), which is a semantic error.
