Merge branch 'ticket28839_v2_squashed'

changes/ticket28839

@@ -0,0 +1,3 @@
+  o Minor features (performance):
+    - Speed up microdesriptor parsing by about 30%, to help
+      improve startup time.  Closes ticket 28839.

src/feature/dirparse/microdesc_parse.c

@@ -177,8 +177,8 @@ microdescs_parse_from_string(const char *s, const char *eos,
                "Relay's onion key had invalid exponent.");
       goto next;
     }
-    router_set_rsa_onion_pkey(tok->key, &md->onion_pkey,
-                              &md->onion_pkey_len);
+    md->onion_pkey = tor_memdup(tok->object_body, tok->object_size);
+    md->onion_pkey_len = tok->object_size;
     crypto_pk_free(tok->key);
 
     if ((tok = find_opt_by_keyword(tokens, K_ONION_KEY_NTOR))) {

src/feature/dirparse/parsecommon.c

@@ -266,7 +266,7 @@ get_next_token(memarea_t *area,
    * attack, a bug, or some other nonsense. */
 #define MAX_LINE_LENGTH (128*1024)
 
-  const char *next, *eol, *obstart;
+  const char *next, *eol;
   size_t obname_len;
   int i;
   directory_token_t *tok;
@@ -352,7 +352,6 @@ get_next_token(memarea_t *area,
   if (!eol || eol-*s<11 || strcmpstart(*s, "-----BEGIN ")) /* No object. */
     goto check_object;
 
-  obstart = *s; /* Set obstart to start of object spec */
   if (eol - *s <= 16 || memchr(*s+11,'\0',eol-*s-16) || /* no short lines, */
       strcmp_len(eol-5, "-----", 5) ||           /* nuls or invalid endings */
       (eol-*s) > MAX_UNPARSED_OBJECT_SIZE) {     /* name too long */
@@ -383,22 +382,26 @@ get_next_token(memarea_t *area,
   if (next - *s > MAX_UNPARSED_OBJECT_SIZE)
     RET_ERR("Couldn't parse object: missing footer or object much too big.");
 
-  if (!strcmp(tok->object_type, "RSA PUBLIC KEY")) { /* If it's a public key */
-    tok->key = crypto_pk_new();
-    if (crypto_pk_read_public_key_from_string(tok->key, obstart, eol-obstart))
-      RET_ERR("Couldn't parse public key.");
-  } else if (!strcmp(tok->object_type, "RSA PRIVATE KEY")) { /* private key */
-    tok->key = crypto_pk_new();
-    if (crypto_pk_read_private_key_from_string(tok->key, obstart, eol-obstart))
-      RET_ERR("Couldn't parse private key.");
-  } else { /* If it's something else, try to base64-decode it */
+  {
     int r;
-    tok->object_body = ALLOC(next-*s); /* really, this is too much RAM. */
-    r = base64_decode(tok->object_body, next-*s, *s, next-*s);
+    size_t maxsize = base64_decode_maxsize(next-*s);
+    tok->object_body = ALLOC(maxsize);
+    r = base64_decode(tok->object_body, maxsize, *s, next-*s);
     if (r<0)
       RET_ERR("Malformed object: bad base64-encoded data");
     tok->object_size = r;
   }
+
+  if (!strcmp(tok->object_type, "RSA PUBLIC KEY")) { /* If it's a public key */
+    tok->key = crypto_pk_asn1_decode(tok->object_body, tok->object_size);
+    if (! tok->key)
+      RET_ERR("Couldn't parse public key.");
+  } else if (!strcmp(tok->object_type, "RSA PRIVATE KEY")) { /* private key */
+    tok->key = crypto_pk_asn1_decode_private(tok->object_body,
+                                             tok->object_size);
+    if (! tok->key)
+      RET_ERR("Couldn't parse private key.");
+  }
   *s = eol;
 
  check_object:

src/feature/dirparse/routerparse.c

@@ -588,8 +588,8 @@ router_parse_entry_from_string(const char *s, const char *end,
              "Relay's onion key had invalid exponent.");
     goto err;
   }
-  router_set_rsa_onion_pkey(tok->key, &router->onion_pkey,
-                            &router->onion_pkey_len);
+  router->onion_pkey = tor_memdup(tok->object_body, tok->object_size);
+  router->onion_pkey_len = tok->object_size;
   crypto_pk_free(tok->key);
 
   if ((tok = find_opt_by_keyword(tokens, K_ONION_KEY_NTOR))) {

src/lib/encoding/binascii.c

@@ -179,6 +179,18 @@ base64_encode_size(size_t srclen, int flags)
   return enclen;
 }
 
+/** Return an upper bound on the number of bytes that might be needed to hold
+ * the data from decoding the base64 string <b>srclen</b>.  This is only an
+ * upper bound, since some part of the base64 string might be padding or
+ * space. */
+size_t
+base64_decode_maxsize(size_t srclen)
+{
+  tor_assert(srclen < INT_MAX / 3);
+
+  return CEIL_DIV(srclen * 3, 4);
+}
+
 /** Internal table mapping 6 bit values to the Base64 alphabet. */
 static const char base64_encode_table[64] = {
   'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H',

src/lib/encoding/binascii.h

@@ -42,6 +42,7 @@ const char *hex_str(const char *from, size_t fromlen);
 
 #define BASE64_ENCODE_MULTILINE 1
 size_t base64_encode_size(size_t srclen, int flags);
+size_t base64_decode_maxsize(size_t srclen);
 int base64_encode(char *dest, size_t destlen, const char *src, size_t srclen,
                   int flags);
 int base64_decode(char *dest, size_t destlen, const char *src, size_t srclen);

src/test/bench.c

@@ -39,6 +39,9 @@
 #include "lib/crypt_ops/digestset.h"
 #include "lib/crypt_ops/crypto_init.h"
 
+#include "feature/dirparse/microdesc_parse.h"
+#include "feature/nodelist/microdesc.h"
+
 #if defined(HAVE_CLOCK_GETTIME) && defined(CLOCK_PROCESS_CPUTIME_ID)
 static uint64_t nanostart;
 static inline uint64_t
@@ -639,6 +642,41 @@ bench_ecdh_p224(void)
 }
 #endif
 
+static void
+bench_md_parse(void)
+{
+  uint64_t start, end;
+  const int N = 100000;
+  // selected arbitrarily
+  const char md_text[] =
+    "@last-listed 2018-12-14 18:14:14\n"
+    "onion-key\n"
+    "-----BEGIN RSA PUBLIC KEY-----\n"
+    "MIGJAoGBAMHkZeXNDX/49JqM2BVLmh1Fnb5iMVnatvZZTLJyedqDLkbXZ1WKP5oh\n"
+    "7ec14dj/k3ntpwHD4s2o3Lb6nfagWbug4+F/rNJ7JuFru/PSyOvDyHGNAuegOXph\n"
+    "3gTGjdDpv/yPoiadGebbVe8E7n6hO+XxM2W/4dqheKimF0/s9B7HAgMBAAE=\n"
+    "-----END RSA PUBLIC KEY-----\n"
+    "ntor-onion-key QgF/EjqlNG1wRHLIop/nCekEH+ETGZSgYOhu26eiTF4=\n"
+    "family $00E9A86E7733240E60D8435A7BBD634A23894098 "
+    "$329BD7545DEEEBBDC8C4285F243916F248972102 "
+    "$69E06EBB2573A4F89330BDF8BC869794A3E10E4D "
+    "$DCA2A3FAE50B3729DAA15BC95FB21AF03389818B\n"
+    "p accept 53,80,443,5222-5223,25565\n"
+    "id ed25519 BzffzY99z6Q8KltcFlUTLWjNTBU7yKK+uQhyi1Ivb3A\n";
+
+  reset_perftime();
+  start = perftime();
+  for (int i = 0; i < N; ++i) {
+    smartlist_t *s = microdescs_parse_from_string(md_text, NULL, 1,
+                                                  SAVED_IN_CACHE, NULL);
+    SMARTLIST_FOREACH(s, microdesc_t *, md, microdesc_free(md));
+    smartlist_free(s);
+  }
+
+  end = perftime();
+  printf("Microdesc parse: %f nsec\n", NANOCOUNT(start, end, N));
+}
+
 typedef void (*bench_fn)(void);
 
 typedef struct benchmark_t {
@@ -666,6 +704,8 @@ static struct benchmark_t benchmarks[] = {
   ENT(ecdh_p256),
   ENT(ecdh_p224),
 #endif
+
+  ENT(md_parse),
   {NULL,NULL,0}
 };
 

src/test/test_parsecommon.c

@@ -300,8 +300,8 @@ test_parsecommon_get_next_token_parse_keys(void *arg)
   tt_int_op(token->tp, OP_EQ, R_IPO_ONION_KEY);
   tt_int_op(token->n_args, OP_EQ, 0);
   tt_str_op(token->object_type, OP_EQ, "RSA PUBLIC KEY");
-  tt_int_op(token->object_size, OP_EQ, 0);
-  tt_assert(!token->object_body);
+  tt_int_op(token->object_size, OP_EQ, 140);
+  tt_assert(token->object_body);
   tt_assert(token->key);
   tt_assert(!token->error);
 
@@ -335,8 +335,8 @@ test_parsecommon_get_next_token_parse_keys(void *arg)
   tt_int_op(token2->tp, OP_EQ, C_CLIENT_KEY);
   tt_int_op(token2->n_args, OP_EQ, 0);
   tt_str_op(token2->object_type, OP_EQ, "RSA PRIVATE KEY");
-  tt_int_op(token2->object_size, OP_EQ, 0);
-  tt_assert(!token2->object_body);
+  tt_int_op(token2->object_size, OP_EQ, 608);
+  tt_assert(token2->object_body);
   tt_assert(token2->key);
   tt_assert(!token->error);
 

src/test/test_util_format.c

@@ -392,10 +392,13 @@ test_util_format_encoded_size(void *arg)
 
     base64_encode(outbuf, sizeof(outbuf), (char *)inbuf, i, 0);
     tt_int_op(strlen(outbuf), OP_EQ, base64_encode_size(i, 0));
+    tt_int_op(i, OP_LE, base64_decode_maxsize(strlen(outbuf)));
+
     base64_encode(outbuf, sizeof(outbuf), (char *)inbuf, i,
                   BASE64_ENCODE_MULTILINE);
     tt_int_op(strlen(outbuf), OP_EQ,
               base64_encode_size(i, BASE64_ENCODE_MULTILINE));
+    tt_int_op(i, OP_LE, base64_decode_maxsize(strlen(outbuf)));
   }
 
  done:
@@ -417,4 +420,3 @@ struct testcase_t util_format_tests[] = {
   { "encoded_size", test_util_format_encoded_size, 0, NULL, NULL },
   END_OF_TESTCASES
 };
-