aboutsummaryrefslogtreecommitdiffstats
path: root/src/cmd/asn1.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/cmd/asn1.cpp')
-rw-r--r--src/cmd/asn1.cpp319
1 files changed, 319 insertions, 0 deletions
diff --git a/src/cmd/asn1.cpp b/src/cmd/asn1.cpp
new file mode 100644
index 000000000..eeda6f809
--- /dev/null
+++ b/src/cmd/asn1.cpp
@@ -0,0 +1,319 @@
+#include "apps.h"
+
+#include <botan/bigint.h>
+#include <botan/der_enc.h>
+#include <botan/ber_dec.h>
+#include <botan/asn1_time.h>
+#include <botan/asn1_str.h>
+#include <botan/oids.h>
+#include <botan/pem.h>
+#include <botan/charset.h>
+using namespace Botan;
+
+#include <sstream>
+#include <stdio.h>
+#include <ctype.h>
+
+// Set this if your terminal understands UTF-8; otherwise output is in Latin-1
+#define UTF8_TERMINAL 1
+
+/*
+ What level the outermost layer of stuff is at. Probably 0 or 1; asn1parse
+ uses 0 as the outermost, while 1 makes more sense to me. 2+ doesn't make
+ much sense at all.
+*/
+#define INITIAL_LEVEL 0
+
+void decode(BER_Decoder&, size_t);
+void emit(const std::string&, size_t, size_t, const std::string& = "");
+std::string type_name(ASN1_Tag);
+
+namespace {
+
+std::string url_encode(const std::vector<byte>& in)
+ {
+ std::ostringstream out;
+
+ for(size_t i = 0; i != in.size(); ++i)
+ {
+ const int c = in[i];
+ if(isprint(c))
+ out << static_cast<char>(c);
+ else
+ out << "%" << std::hex << static_cast<char>(c) << std::dec;
+ }
+ return out.str();
+ }
+
+}
+
+int asn1_main(int argc, char* argv[])
+ {
+ if(argc != 2)
+ {
+ printf("Usage: %s <file>\n", argv[0]);
+ return 1;
+ }
+
+ try {
+ DataSource_Stream in(argv[1]);
+
+ if(!PEM_Code::matches(in))
+ {
+ BER_Decoder decoder(in);
+ decode(decoder, INITIAL_LEVEL);
+ }
+ else
+ {
+ std::string label; // ignored
+ BER_Decoder decoder(PEM_Code::decode(in, label));
+ decode(decoder, INITIAL_LEVEL);
+ }
+
+ }
+ catch(std::exception& e)
+ {
+ printf("%s\n", e.what());
+ return 1;
+ }
+ return 0;
+ }
+
+void decode(BER_Decoder& decoder, size_t level)
+ {
+ BER_Object obj = decoder.get_next_object();
+
+ while(obj.type_tag != NO_OBJECT)
+ {
+ const ASN1_Tag type_tag = obj.type_tag;
+ const ASN1_Tag class_tag = obj.class_tag;
+ const size_t length = obj.value.size();
+
+ /* hack to insert the tag+length back in front of the stuff now
+ that we've gotten the type info */
+ DER_Encoder encoder;
+ encoder.add_object(type_tag, class_tag, obj.value);
+ std::vector<byte> bits = encoder.get_contents_unlocked();
+
+ BER_Decoder data(bits);
+
+ if(class_tag & CONSTRUCTED)
+ {
+ BER_Decoder cons_info(obj.value);
+ if(type_tag == SEQUENCE)
+ {
+ emit("SEQUENCE", level, length);
+ decode(cons_info, level+1);
+ }
+ else if(type_tag == SET)
+ {
+ emit("SET", level, length);
+ decode(cons_info, level+1);
+ }
+ else
+ {
+ std::string name;
+
+ if((class_tag & APPLICATION) || (class_tag & CONTEXT_SPECIFIC))
+ {
+ name = "cons [" + std::to_string(type_tag) + "]";
+
+ if(class_tag & APPLICATION)
+ name += " appl";
+ if(class_tag & CONTEXT_SPECIFIC)
+ name += " context";
+ }
+ else
+ name = type_name(type_tag) + " (cons)";
+
+ emit(name, level, length);
+ decode(cons_info, level+1);
+ }
+ }
+ else if((class_tag & APPLICATION) || (class_tag & CONTEXT_SPECIFIC))
+ {
+#if 0
+ std::vector<byte> bits;
+ data.decode(bits, type_tag);
+
+ try
+ {
+ BER_Decoder inner(bits);
+ decode(inner, level + 1);
+ }
+ catch(...)
+ {
+ emit("[" + std::to_string(type_tag) + "]", level, length,
+ url_encode(bits));
+ }
+#else
+ emit("[" + std::to_string(type_tag) + "]", level, length,
+ url_encode(bits));
+#endif
+ }
+ else if(type_tag == OBJECT_ID)
+ {
+ OID oid;
+ data.decode(oid);
+
+ std::string out = OIDS::lookup(oid);
+ if(out != oid.as_string())
+ out += " [" + oid.as_string() + "]";
+
+ emit(type_name(type_tag), level, length, out);
+ }
+ else if(type_tag == INTEGER || type_tag == ENUMERATED)
+ {
+ BigInt number;
+
+ if(type_tag == INTEGER)
+ data.decode(number);
+ else if(type_tag == ENUMERATED)
+ data.decode(number, ENUMERATED, class_tag);
+
+ std::vector<byte> rep;
+
+ /* If it's small, it's probably a number, not a hash */
+ if(number.bits() <= 16)
+ rep = BigInt::encode(number, BigInt::Decimal);
+ else
+ rep = BigInt::encode(number, BigInt::Hexadecimal);
+
+ std::string str;
+ for(size_t i = 0; i != rep.size(); ++i)
+ str += static_cast<char>(rep[i]);
+
+ emit(type_name(type_tag), level, length, str);
+ }
+ else if(type_tag == BOOLEAN)
+ {
+ bool boolean;
+ data.decode(boolean);
+ emit(type_name(type_tag),
+ level, length, (boolean ? "true" : "false"));
+ }
+ else if(type_tag == NULL_TAG)
+ {
+ emit(type_name(type_tag), level, length);
+ }
+ else if(type_tag == OCTET_STRING)
+ {
+ std::vector<byte> bits;
+ data.decode(bits, type_tag);
+
+ try
+ {
+ BER_Decoder inner(bits);
+ decode(inner, level + 1);
+ }
+ catch(...)
+ {
+ emit(type_name(type_tag), level, length,
+ url_encode(bits));
+ }
+ }
+ else if(type_tag == BIT_STRING)
+ {
+ std::vector<byte> bits;
+ data.decode(bits, type_tag);
+
+ std::vector<bool> bit_set;
+
+ for(size_t i = 0; i != bits.size(); ++i)
+ for(size_t j = 0; j != 8; ++j)
+ bit_set.push_back((bool)((bits[bits.size()-i-1] >> (7-j)) & 1));
+
+ std::string bit_str;
+ for(size_t i = 0; i != bit_set.size(); ++i)
+ {
+ bool the_bit = bit_set[bit_set.size()-i-1];
+
+ if(!the_bit && bit_str.size() == 0)
+ continue;
+ bit_str += (the_bit ? "1" : "0");
+ }
+
+ emit(type_name(type_tag), level, length, bit_str);
+ }
+ else if(type_tag == PRINTABLE_STRING ||
+ type_tag == NUMERIC_STRING ||
+ type_tag == IA5_STRING ||
+ type_tag == T61_STRING ||
+ type_tag == VISIBLE_STRING ||
+ type_tag == UTF8_STRING ||
+ type_tag == BMP_STRING)
+ {
+ ASN1_String str;
+ data.decode(str);
+ if(UTF8_TERMINAL)
+ emit(type_name(type_tag), level, length,
+ Charset::transcode(str.iso_8859(),
+ LATIN1_CHARSET, UTF8_CHARSET));
+ else
+ emit(type_name(type_tag), level, length, str.iso_8859());
+ }
+ else if(type_tag == UTC_TIME || type_tag == GENERALIZED_TIME)
+ {
+ X509_Time time;
+ data.decode(time);
+ emit(type_name(type_tag), level, length, time.readable_string());
+ }
+ else
+ fprintf(stderr, "Unknown tag: class=%02X, type=%02X\n",
+ class_tag, type_tag);
+
+ obj = decoder.get_next_object();
+ }
+ }
+
+void emit(const std::string& type, size_t level, size_t length,
+ const std::string& value)
+ {
+ const size_t LIMIT = 128;
+ const size_t BIN_LIMIT = 64;
+
+ int written = 0;
+ written += printf(" d=%2d, l=%4d: ", (int)level, (int)length);
+ for(size_t i = INITIAL_LEVEL; i != level; ++i)
+ written += printf(" ");
+ written += printf("%s ", type.c_str());
+
+ bool should_skip = false;
+ if(value.length() > LIMIT) should_skip = true;
+ if((type == "OCTET STRING" || type == "BIT STRING") &&
+ value.length() > BIN_LIMIT)
+ should_skip = true;
+
+ if(value != "" && !should_skip)
+ {
+ if(written % 2 == 0) printf(" ");
+ while(written < 50) written += printf(" ");
+ printf(":%s\n", value.c_str());
+ }
+ else
+ printf("\n");
+ }
+
+std::string type_name(ASN1_Tag type)
+ {
+ if(type == PRINTABLE_STRING) return "PRINTABLE STRING";
+ if(type == NUMERIC_STRING) return "NUMERIC STRING";
+ if(type == IA5_STRING) return "IA5 STRING";
+ if(type == T61_STRING) return "T61 STRING";
+ if(type == UTF8_STRING) return "UTF8 STRING";
+ if(type == VISIBLE_STRING) return "VISIBLE STRING";
+ if(type == BMP_STRING) return "BMP STRING";
+
+ if(type == UTC_TIME) return "UTC TIME";
+ if(type == GENERALIZED_TIME) return "GENERALIZED TIME";
+
+ if(type == OCTET_STRING) return "OCTET STRING";
+ if(type == BIT_STRING) return "BIT STRING";
+
+ if(type == ENUMERATED) return "ENUMERATED";
+ if(type == INTEGER) return "INTEGER";
+ if(type == NULL_TAG) return "NULL";
+ if(type == OBJECT_ID) return "OBJECT";
+ if(type == BOOLEAN) return "BOOLEAN";
+ return "(UNKNOWN)";
+ }