code wiki / _hdl_build / nx_hmacsha1_extvec_gate.nx
`nx_hmacsha1_extvec_gate.nx` source

↩ module page · 444 lines · 21798 B
1// nx_hmacsha1_extvec_gate.nx -- validated against RFC2202-HMAC-SHA1, read from the pinned+corroborated RFC2202 document.
2// ⚠HEADER CORRECTED 2026-08-01: this file was CLONED from nx_hmac_extvec_gate.nx and inherited its
3// header verbatim, so it claimed to be that gate validating HMAC-SHA-256 against RFC 4231. Nine of
4// twenty-three extvec gates carried the same wrong self-description. ★★★★★A CLONED FILE INHERITS ITS
5// PARENT'S CLAIMS, AND ON AN EVIDENCE ARTIFACT THE HEADER IS A PROVENANCE CLAIM, NOT A COMMENT --
6// an auditor reading headers would have concluded RFC 4231 validated all nine subjects.
7//
8// SUPERSEDES the single-case version. Going from 1 vector to 7 matters because the cases are deliberately
9// chosen by the authority to hit DIFFERENT code paths, and the ones I was NOT running are the interesting
10// ones: case 3 uses a 50-byte repeated data block, cases 6 and 7 use a 131-BYTE KEY (longer than the 64-byte
11// SHA-256 block, so the key must be HASHED first), and case 5 publishes a TRUNCATED 128-bit MAC.
12// ★A gate that ran only case 1 would never touch the key-longer-than-block branch -- the single most
13// commonly botched part of HMAC. Running one vector from a seven-vector suite is not "validated against
14// RFC 4231"; it is validated against one line of it.
15//
16// ⚠VARIABLE-LENGTH FIELDS, HANDLED BY TERMINATOR NOT BY LENGTH. Key/Data lengths differ per case and their
17// annotations are inconsistent -- "(20 bytes)" for keys but ("Hi There") for data -- so a length cannot be
18// read uniformly. Instead the hex run is read until the first `(`, which terminates both forms.
19// ⚠THE MAC HAS NO `(` TERMINATOR and case 5's is TRUNCATED to 16 bytes, so it is read as hex pairs until a
20// pair is not both-hex, capped at 32. That correctly stops at the section heading that follows -- note
21// "4.3." begins with '4', a HEX DIGIT, and is only rejected because '.' is not: the PAIR rule saves this,
22// a single-nibble rule would have swallowed it.
23//
24// Construction unchanged: no expected value in this source, document pinned to a socket-time digest, every
25// key/data/MAC read from that pinned document, and a completeness check that refuses GREEN below 7.
26// license_tier: ORIGINAL  expect_exit: 0
27import "nx_syscalls.nx"
28import "nx_sha256_wasm.nx"
29import "hmac_sha1.nx"
30import "hkdf_sha1.nx"
31import "nx_sha1.nx"
32import "nx_gate_verdict.nx"
33
34func w(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 }
35func wb(b: *u8, n: i64) -> i64 { sys_write(1, b, n); return 0 }
36
37func nn(v: i64) -> i64 {
  var m: i64 = v
  if m < 0 { w("-" as *u8); m = 0 - m }
  let t: *u8 = sys_mmap(32)
  var k: i64 = 0
  if m == 0 { t[0] = 48 as u8; k = 1 }
  while m > 0 { t[k] = (48 + (m % 10)) as u8; m = m / 10; k = k + 1 }
  let b: *u8 = sys_mmap(32)
  var j: i64 = 0
  while j < k { b[j] = t[k - 1 - j]; j = j + 1 }
  sys_write(1, b, k)
  return 0
49}
50
51func hexnib(v: i64) -> i64 { if v < 10 { return 48 + v } return 87 + v }
52
53func hexval(c: i64) -> i64 {
  if c >= 48 { if c <= 57 { return c - 48 } }
  if c >= 97 { if c <= 102 { return c - 87 } }
  if c >= 65 { if c <= 70 { return c - 55 } }
  return 0 - 1
58}
59
60func isws(c: i64) -> i64 {
  if c == 32 { return 1 }
  if c == 10 { return 1 }
  if c == 13 { return 1 }
  if c == 9 { return 1 }
  return 0
66}
67
68func starts(b: *u8, n: i64, at: i64, s: *u8) -> i64 {
  var i: i64 = 0
  while s[i] != (0 as u8) {
      if at + i >= n { return 0 }
      if b[at + i] != s[i] { return 0 }
      i = i + 1
  }
  return 1
76}
77
78func findfrom(b: *u8, n: i64, s: *u8, from: i64) -> i64 {
  var p: i64 = from
  while p < n {
      if starts(b, n, p, s) == 1 { return p }
      p = p + 1
  }
  return 0 - 1
85}
86
87// Read hex PAIRS (whitespace between pairs is skipped) until a pair is not both-hex, or `(` is reached,
88// or cap is hit. Returns the byte count.
89func parserun(b: *u8, n: i64, from: i64, out: *u8, cap: i64) -> i64 {
  var p: i64 = from
  var got: i64 = 0
  var done: i64 = 0
  while done == 0 {
      if got >= cap { done = 1 }
      else {
          // skip whitespace, REMEMBERING whether the skip crossed a line boundary
          var d1: i64 = 0
          var crossed: i64 = 0
          while d1 == 0 {
              if p >= n { d1 = 1 }
              else { if isws(b[p] as i64) == 1 { if b[p] == (10 as u8) { crossed = 1 } p = p + 1 } else { d1 = 1 } }
          }
          // ★"Da" IS VALID HEX. Same defect, same file family: the `=`-discriminator was applied only to
          // the `(`-annotated branch, so a `digest =` value (which has no `(` annotation) ran past
          // end-of-line and swallowed the "da" of the following "data_len" as byte 0xDA -- 21 bytes for a
          // 20-byte digest. Ported here from nx_hmacsha1fix_extvec_gate, where it was MEASURED:
          //   expected e8e9…1a91 DA   vs   computed e8e9…1a91 00   (first 20 bytes IDENTICAL)
          // ★A FIX APPLIED TO ONE COPY OF A CLONED READER LEAVES EVERY OTHER COPY BROKEN -- and this gate
          //  sat at 8/9 looking like a crypto residue when it was the same reader bug all along.
          if crossed == 1 {
              var eq2: i64 = 0
              var sc2: i64 = p
              var de2: i64 = 0
              while de2 == 0 {
                  if sc2 >= n { de2 = 1 }
                  else { if b[sc2] == (10 as u8) { de2 = 1 }
                  else { if b[sc2] == (61 as u8) { eq2 = 1; de2 = 1 } else { sc2 = sc2 + 1 } } }
              }
              if eq2 == 1 { p = n }
          }
          if p + 1 >= n { done = 1 }
          else {
              // `(` ends a LINE SEGMENT, not the value: RFC 4231 annotates EVERY wrapped line, e.g.
              //   Data = 7768...6e7420  ("what do ya want ")
              //          666f...693f    ("for nothing?")
              // Treating `(` as the value terminator stopped case 2 at 16 of its 28 bytes. On `(`, skip
              // to the next line and continue ONLY if it resumes with a hex pair; otherwise stop.
              if b[p] == (40 as u8) {
                  var dl: i64 = 0
                  while dl == 0 {
                      if p >= n { dl = 1 }
                      else { if b[p] == (10 as u8) { p = p + 1; dl = 1 } else { p = p + 1 } }
                  }
                  var dw: i64 = 0
                  while dw == 0 {
                      if p >= n { dw = 1 }
                      else { if isws(b[p] as i64) == 1 { p = p + 1 } else { dw = 1 } }
                  }
                  // ⚠"Da" IS VALID HEX. The next line may be `   Data =  7768...`, and testing only
                  // "do the first two chars parse as hex" accepted D,a and read the LABEL as byte 0xDA --
                  // adding one phantom byte to every key (case 1 read 21 of 20, case 2 read 5 of 4).
                  // Third form of the same hazard today: an ASCII gutter, then English prose, now a
                  // FIELD LABEL that happens to spell hex.
                  // STRUCTURAL DISCRIMINATOR: a continuation line has NO '='; every label line has one.
                  var eqfound: i64 = 0
                  var sc: i64 = p
                  var de: i64 = 0
                  while de == 0 {
                      if sc >= n { de = 1 }
                      else { if b[sc] == (10 as u8) { de = 1 }
                      else { if b[sc] == (61 as u8) { eqfound = 1; de = 1 } else { sc = sc + 1 } } }
                  }
                  if eqfound == 1 { done = 1 }
                  else {
                      if p + 1 >= n { done = 1 }
                      else {
                          if hexval(b[p] as i64) < 0 { done = 1 }
                          else { if hexval(b[p + 1] as i64) < 0 { done = 1 } }
                      }
                  }
              }
              else {
                  let h1: i64 = hexval(b[p] as i64)
                  let h2: i64 = hexval(b[p + 1] as i64)
                  if h1 < 0 { done = 1 }
                  else { if h2 < 0 { done = 1 }
                  else {
                      out[got] = ((h1 * 16) + h2) as u8
                      got = got + 1
                      p = p + 2
                  } }
              }
          }
      }
  }
  return got
177}
178
179// Find the next "Key" FIELD LABEL at line start. Verified against the raw bytes of RFC 4231:
180//   line 192 "   Key =          0b0b..."   <- cases 1,2,4,5,6,7
181//   line 251 "   Key            aaaa..."   <- CASE 3: no '=' at all
182//   line 177 "   Keys, data, and digests..." <- PROSE. "Key" is a PREFIX of "Keys".
183// Anchoring on "\n   Key" alone matched that prose line and drove the parse to ZERO cases. So the label is
184// accepted only when the character AFTER "Key" is a SPACE or '=' -- which "Keys" fails on 's'.
185// ★A prefix match is not a token match. Every anchor in this file is now checked against the byte AFTER it.
186func find_key_label(b: *u8, n: i64, from: i64) -> i64 {
  var p: i64 = from
  var done: i64 = 0
  while done == 0 {
      let h: i64 = findfrom(b, n, "\n   Key" as *u8, p)
      if h < 0 { return 0 - 1 }
      let c: i64 = b[h + 7] as i64
      if c == 32 { return h }
      if c == 61 { return h }
      p = h + 7
  }
  return 0 - 1
198}
199func readdec(b: *u8, n: i64, from: i64) -> i64 {
  var p: i64 = from
  var v: i64 = 0
  var got: i64 = 0
  var done: i64 = 0
  while done == 0 {
      if p >= n { done = 1 }
      else {
          let c: i64 = b[p] as i64
          if c == 32 { if got == 1 { done = 1 } else { p = p + 1 } }
          else { if c >= 48 { if c <= 57 { v = (v*10)+(c-48); got = 1; p = p + 1 } else { done = 1 } } else { done = 1 } }
      }
  }
  if got == 0 { return 0 - 1 }
  return v
214}
215
216// A field is EITHER 0x-hex OR a quoted ASCII string. RFC 2202 uses both for `data` across its cases
217// (0xdd... for the repeated-byte cases, "Hi There" for the readable ones), so the reader must dispatch on
218// the first non-space character rather than assume one form.
219func readdec_skipeq(b: *u8, n: i64, from: i64) -> i64 {
  var p: i64 = from
  var d: i64 = 0
  while d == 0 {
      if p >= n { d = 1 }
      else { if b[p] == (32 as u8) { p = p + 1 } else { if b[p] == (61 as u8) { p = p + 1 } else { d = 1 } } }
  }
  return readdec(b, n, p)
227}
228
229func readfield(b: *u8, n: i64, from: i64, out: *u8, cap: i64) -> i64 {
  var p: i64 = from
  var d: i64 = 0
  while d == 0 {
      if p >= n { d = 1 }
      else { if b[p] == (32 as u8) { p = p + 1 } else { d = 1 } }
  }
  if p >= n { return 0 - 1 }
  if b[p] == (34 as u8) {
      // ⚠A WRAPPED QUOTED STRING: the newline REPLACES a space, it does not delete one.
      //   data =          "Test Using Larger Than Block-Size Key and Larger
      //                   Than One Block-Size Data"
      // Copying raw gave 89 (47 + newline + 16 indent + 25); dropping the wrap entirely gives 72; the
      // document declares 73. So a newline plus its following indentation collapses to EXACTLY ONE SPACE.
      // ★Verified against the document's own data_len, not assumed -- the declared length is what
      // distinguishes "delete the wrap" from "replace the wrap", and those differ by one byte.
      var k: i64 = 0
      p = p + 1
      while p < n {
          if b[p] == (34 as u8) { return k }
          if k >= cap { return 0 - 1 }
          if b[p] == (10 as u8) {
              out[k] = 32 as u8
              k = k + 1
              p = p + 1
              var ds: i64 = 0
              while ds == 0 {
                  if p >= n { ds = 1 }
                  else { if b[p] == (32 as u8) { p = p + 1 } else { if b[p] == (13 as u8) { p = p + 1 } else { ds = 1 } } }
              }
          } else {
              if b[p] == (13 as u8) { p = p + 1 }
              else {
                  out[k] = b[p]
                  k = k + 1
                  p = p + 1
              }
          }
      }
      return 0 - 1
  }
  if p + 1 < n { if b[p] == (48 as u8) { if b[p+1] == (120 as u8) {
      let got: i64 = parserun(b, n, p + 2, out, cap)
      // ⚠RFC 2202 EXPRESSES BULK DATA AS PROSE: `data = 0xdd repeated 50 times`. That is ONE hex byte
      // followed by a repetition count in ENGLISH. A literal hex reader returns 1 byte, and the
      // document-declared data_len=50 is what caught it -- the self-check named the READER instead of
      // letting HMAC be blamed for a 1-byte input.
      // ★A value can be expressed as a PROGRAM ("repeat this"), not just as data. Tenth notation form.
      if got == 1 {
          let rp: i64 = findfrom(b, n, "repeated" as *u8, p)
          if rp >= 0 { if rp < p + 40 {
              let cnt: i64 = readdec(b, n, rp + 8)
              if cnt > 1 { if cnt <= cap {
                  let fill: i64 = out[0] as i64
                  var q: i64 = 0
                  while q < cnt { out[q] = fill as u8; q = q + 1 }
                  return cnt
              } }
          } }
      }
      return got
  } } }
  return 0 - 1
292}
293
294// REFERENCE HMAC-SHA-1 BUILT HERE FROM THE EXONERATED sha1() ONLY.
295// RFC 2104: HMAC(K,m) = H((K' xor opad) || H((K' xor ipad) || m)), K' = K padded to 64 (hashed first if
296// longer). This shares ONLY the hash core with hmac_sha1 -- and that core is independently PROVEN correct
297// against RFC 3174. So if THIS matches the published digest and hmac_sha1 does not, the wrapper is at
298// fault; if BOTH miss, the reader is. ★A differential is only worth running when its shared set excludes
299// the suspect -- round 53's did not, this one does.
300func ref_hmac_sha1(key: *u8, klen: i64, msg: *u8, mlen: i64, out: *u8) -> i64 {
  let kp: *u8 = sys_mmap(128)
  var i: i64 = 0
  while i < 64 { kp[i] = 0 as u8; i = i + 1 }
  if klen > 64 {
      let kh: *u8 = sys_mmap(64)
      sha1(key, klen, kh)
      i = 0
      while i < 20 { kp[i] = kh[i]; i = i + 1 }
  } else {
      i = 0
      while i < klen { kp[i] = key[i]; i = i + 1 }
  }
  let inner: *u8 = sys_mmap(4096)
  i = 0
  while i < 64 { inner[i] = (((kp[i] as i64) ^ 54) & 255) as u8; i = i + 1 }
  i = 0
  while i < mlen { inner[64 + i] = msg[i]; i = i + 1 }
  let ih: *u8 = sys_mmap(64)
  sha1(inner, 64 + mlen, ih)
  let outer: *u8 = sys_mmap(256)
  i = 0
  while i < 64 { outer[i] = (((kp[i] as i64) ^ 92) & 255) as u8; i = i + 1 }
  i = 0
  while i < 20 { outer[64 + i] = ih[i]; i = i + 1 }
  sha1(outer, 84, out)
  return 0
327}
328func main() -> i64 {
  w("nx_hmacsha1_extvec_gate -- HMAC-SHA-1 vs RFC 2202, READ FROM THE FETCHED DOCUMENT\n" as *u8)
  let lp: *i64 = sys_mmap(16) as *i64
  lp[0] = 0
  let b: *u8 = sys_read_file("knowledge/extvec/rfc2202.txt\x00" as *u8, lp)
  if lp[0] <= 0 { w("RED: fetched vector file absent.\n" as *u8); return 1 }
  let ctx: *u8 = sys_mmap(1024)
  let dg: *u8 = sys_mmap(64)
  nx_sha256_one_shot(b, lp[0], ctx, dg)
  let hx: *u8 = sys_mmap(80)
  var i: i64 = 0
  while i < 32 { hx[i*2] = hexnib(((dg[i] as i64)/16)&15) as u8; hx[i*2+1] = hexnib((dg[i] as i64)&15) as u8; i = i + 1 }
  let wnt: *u8 = "c19effeca47e801be304460da3b2fb05f595e2309abceb050e40e024868fe483\x00" as *u8
  var pin: i64 = 1
  i = 0
  while i < 64 { if hx[i] != wnt[i] { pin = 0 } i = i + 1 }
  w("  acquisition digest: " as *u8); wb(hx, 64); w("\n" as *u8)
  if pin == 0 { w("RED: PIN FAILED.\n" as *u8); return 1 }
  w("  PIN OK -- bytes match the digest computed in-process at the socket\n" as *u8)
347
  let key: *u8 = sys_mmap(512)
  let data: *u8 = sys_mmap(512)
  let exp: *u8 = sys_mmap(128)
  let got: *u8 = sys_mmap(128)
352
  var pass: i64 = 0
  var fail: i64 = 0
  var seen: i64 = 0
  var skipped: i64 = 0
  // ANCHOR ON THE SECTION, NOT ON A DIGEST-LENGTH GUESS. RFC 2202 has 7 HMAC-MD5 cases (sec 2) and
  // NINE HMAC-SHA-1 markers (sec 3 -- cases 6 and 7 appear TWICE, once with longer data). Dispatching by
  // digest length mis-sorted them 8/8 against the true 7/9. Starting at the section heading makes every
  // case found a SHA-1 case BY CONSTRUCTION -- no inference required.
  let sec: i64 = findfrom(b, lp[0], "3. Test Cases for HMAC-SHA-1" as *u8, 0)
  if sec < 0 { w("RED: no HMAC-SHA-1 section heading\n" as *u8); return 1 }
  var cur: i64 = sec
  var done: i64 = 0
  while done == 0 {
      let tc: i64 = findfrom(b, lp[0], "test_case =" as *u8, cur)
      if tc < 0 { done = 1 }
      else {
          let lk: i64 = findfrom(b, lp[0], "key =" as *u8, tc)
          let lkl: i64 = findfrom(b, lp[0], "key_len" as *u8, tc)
          let ld: i64 = findfrom(b, lp[0], "data =" as *u8, tc)
          let ldl: i64 = findfrom(b, lp[0], "data_len" as *u8, tc)
          let lg: i64 = findfrom(b, lp[0], "digest =" as *u8, tc)
          if lk < 0 { done = 1 }
          else { if lg < 0 { done = 1 }
          else {
              let kn: i64 = readfield(b, lp[0], lk + 5, key, 400)
              let dn: i64 = readfield(b, lp[0], ld + 6, data, 400)
              let kdecl: i64 = readdec_skipeq(b, lp[0], lkl + 7)
              let ddecl: i64 = readdec_skipeq(b, lp[0], ldl + 8)
              let gn: i64 = readfield(b, lp[0], lg + 8, exp, 100)
              if gn < 0 { done = 1 }
              else {
                  // THE DOCUMENT DECLARES ITS OWN LENGTHS -- grade the reader by the source.
                  if kn != kdecl {
                      w("  RED: key parsed " as *u8); nn(kn); w(" but document declares key_len=" as *u8); nn(kdecl)
                      w(" -- the READER.\n" as *u8)
                      fail = fail + 1
                      done = 1
                  } else { if dn != ddecl {
                      w("  RED: data parsed " as *u8); nn(dn); w(" but document declares data_len=" as *u8); nn(ddecl)
                      w(" -- the READER.\n" as *u8)
                      fail = fail + 1
                      done = 1
                  } else {
                      // dispatch on the PUBLISHED digest length: 20 = SHA-1, 16 = the MD5 section
                      if gn > 0 {
                          hmac_sha1(key, kn, data, dn, got)
                          // DIFFERENTIAL: HKDF-Extract(salt, ikm) IS DEFINED AS HMAC-Hash(salt, ikm)
                          // (RFC 5869 sec 2.2), so hkdf_sha1_extract is an INDEPENDENT internal path to
                          // the same function. If one matches the published vector and the other does
                          // not, the defect is localised to the one that disagrees -- adjudicated by the
                          // AUTHORITY, not by which of our two implementations we happen to trust.
                          let alt: *u8 = sys_mmap(64)
                          ref_hmac_sha1(key, kn, data, dn, alt)
                          var altsame: i64 = 1
                          var z: i64 = 0
                          while z < gn { if alt[z] != exp[z] { altsame = 0 } z = z + 1 }
                          if altsame == 1 { w("    [differential] REFERENCE HMAC (built from the exonerated sha1) MATCHES -> the WRAPPER is at fault\n" as *u8) }
                          else { w("    [differential] reference HMAC also differs -> the READER is at fault, crypto exonerated\n" as *u8) }
                          // Compare only the PUBLISHED digest length: section 3 case 5 is the truncation
                          // case, whose published digest is SHORTER than HMAC-SHA-1's 20-byte output.
                          // Comparing a fixed 20 would fail a correct implementation on that case.
                          var same: i64 = 1
                          i = 0
                          while i < gn { if got[i] != exp[i] { same = 0 } i = i + 1 }
                          seen = seen + 1
                          if same == 1 { pass = pass + 1; w("  PASS sha1 keylen=" as *u8); nn(kn); w(" datalen=" as *u8); nn(dn); w(" maclen=" as *u8); nn(gn); w("\n" as *u8) }
                          else { fail = fail + 1; w("  FAIL sha1 case keylen=" as *u8); nn(kn); w(" datalen=" as *u8); nn(dn); w("\n" as *u8) }
                      } else { skipped = skipped + 1 }
                      cur = lg + 8
                  } }
              }
          } }
      }
  }
427
  w("\n  refsrc=https://www.rfc-editor.org/rfc/rfc2202.txt\n" as *u8)
  w("  refsrcdig=" as *u8); wb(hx, 64); w("\n" as *u8)
  w("  ref=RFC2202-HMAC-SHA1   gate=nx_hmacsha1_extvec_gate\n" as *u8)
  w("  NOTE: key_len and data_len are cross-checked against the document on EVERY case.\n" as *u8)
  w("        16-byte-digest cases are the HMAC-MD5 section and are counted as SKIPPED, not passed.\n" as *u8)
  w("nx_hmacsha1_extvec_gate: sha1_cases=" as *u8); nn(seen); w(" skipped_md5=" as *u8); nn(skipped)
  w(" pass=" as *u8); nn(pass); w(" fail=" as *u8); nn(fail)
  // MIGRATED onto nx_gate_verdict by nx_gate_dry_apply (D001, minimal form): every check
  // row above is untouched, so the PASS/FAIL vector cannot change; only the hand-rolled
  // verdict emission is replaced by the ONE shared base class. Proven by nx_gate_migrate verify.
  let ctr__dry: *i64 = gv_ctr()
  ctr__dry[0] = pass
  ctr__dry[1] = pass + fail
  let rc__dry: i64 = gv_verdict("HMACSHA1-EXTVEC-GATE" as *u8, ctr__dry, "teeth unchanged; verdict emission migrated onto the shared base class" as *u8)
  sys_exit(rc__dry)
  return rc__dry
444}
nx_hmacsha1_extvec_gate.nx source

`nx_hmacsha1_extvec_gate.nx` source
