summaryrefslogtreecommitdiffstats
path: root/tools/make-enterprises.py
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-10 20:34:10 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-10 20:34:10 +0000
commite4ba6dbc3f1e76890b22773807ea37fe8fa2b1bc (patch)
tree68cb5ef9081156392f1dd62a00c6ccc1451b93df /tools/make-enterprises.py
parentInitial commit. (diff)
downloadwireshark-e4ba6dbc3f1e76890b22773807ea37fe8fa2b1bc.tar.xz
wireshark-e4ba6dbc3f1e76890b22773807ea37fe8fa2b1bc.zip
Adding upstream version 4.2.2.upstream/4.2.2
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'tools/make-enterprises.py')
-rwxr-xr-xtools/make-enterprises.py196
1 files changed, 196 insertions, 0 deletions
diff --git a/tools/make-enterprises.py b/tools/make-enterprises.py
new file mode 100755
index 0000000..1b2b2d0
--- /dev/null
+++ b/tools/make-enterprises.py
@@ -0,0 +1,196 @@
+#!/usr/bin/env python3
+# create the enterprises.c file from
+# https://www.iana.org/assignments/enterprise-numbers/enterprise-numbers
+# or an offline copy
+#
+# Copyright 2022 by Moshe Kaplan
+# Based on make-sminmpec.pl by Gerald Combs
+#
+# Wireshark - Network traffic analyzer
+# By Gerald Combs <gerald@wireshark.org>
+# Copyright 2004 Gerald Combs
+#
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+import os
+import argparse
+import re
+import urllib.request
+
+
+ENTERPRISES_CFILE = os.path.join('epan', 'enterprises.c')
+
+ENTERPRISE_NUMBERS_URL = "https://www.iana.org/assignments/enterprise-numbers/enterprise-numbers"
+
+DECIMAL_PATTERN = r"^(\d+)"
+# up to three spaces because of formatting errors in the source
+ORGANIZATION_PATTERN = r"^ ?(\S.*)"
+FORMERLY_PATTERN = r" \(((formerly|previously) .*)\)"
+
+
+LOOKUP_FUNCTION = r"""
+const char* global_enterprises_lookup(uint32_t value)
+{
+ if (value > table.max_idx) {
+ return NULL;
+ }
+ else return table.values[value];
+}
+"""
+
+DUMP_FUNCTION = r"""
+void global_enterprises_dump(FILE *fp)
+{
+ for (size_t idx = 0; idx <= table.max_idx; idx++) {
+ if (table.values[idx] != NULL) {
+ fprintf(fp, "%zu\t%s\n", idx, table.values[idx]);
+ }
+ }
+}
+"""
+
+# This intermediate format is no longer written to a file - returned as string
+def generate_enterprise_entries(file_content):
+ # We only care about the "Decimal" and "Organization",
+ # not the contact or email
+ org_lines = []
+ last_updated = ""
+ end_seen = False
+ for line in file_content.splitlines():
+ decimal_match = re.match(DECIMAL_PATTERN, line)
+ if decimal_match:
+ decimal = decimal_match.group(0)
+ elif re.match(ORGANIZATION_PATTERN, line):
+ organization = line.strip()
+ if organization.lower() == "unassigned":
+ continue
+ organization = re.sub(FORMERLY_PATTERN, r"\t# \1", organization)
+ org_lines += [decimal + "\t" + organization]
+ elif "last updated" in line.lower():
+ last_updated = line
+ elif "end of document" in line.lower():
+ end_seen = True
+
+ if not end_seen:
+ raise Exception('"End of Document" not found. Truncated source file?')
+
+ last_updated_line = "/* " + last_updated + " */\n\n"
+ output = "\n".join(org_lines) + "\n"
+ return (output,last_updated_line)
+
+class CFile:
+ def __init__(self, filename, last_updated_line):
+ self.filename = filename
+ self.f = open(filename, 'w')
+ self.mappings = {}
+ self.highest_num = 0
+
+ # Write file header
+ self.f.write('/* ' + os.path.basename(self.filename) + '\n')
+ self.f.write(' *\n')
+ self.f.write(' * Wireshark - Network traffic analyzer\n')
+ self.f.write(' * By Gerald Combs <gerald@wireshark.org>\n')
+ self.f.write(' * Copyright 1998 Gerald Combs\n')
+ self.f.write(' *\n')
+ self.f.write(' * Do not edit - this file is automatically generated\n')
+ self.f.write(' * SPDX-License-Identifier: GPL-2.0-or-later\n')
+ self.f.write(' */\n\n')
+ self.f.write(last_updated_line)
+
+ # Include header files
+ self.f.write('#include "config.h"\n\n')
+ self.f.write('#include <stddef.h>\n')
+ self.f.write('#include "enterprises.h"\n')
+ self.f.write('\n\n')
+
+ def __del__(self):
+ self.f.write('typedef struct\n')
+ self.f.write('{\n')
+ self.f.write(' uint32_t max_idx;\n')
+ self.f.write(' const char* values[' + str(self.highest_num+1) + '];\n')
+ self.f.write('} global_enterprises_table_t;\n\n')
+
+ # Write static table
+ self.f.write('static global_enterprises_table_t table =\n')
+ self.f.write('{\n')
+ # Largest index
+ self.f.write(' ' + str(self.highest_num) + ',\n')
+ self.f.write(' {\n')
+ # Entries (read from dict)
+ for n in range(0, self.highest_num+1):
+ if n not in self.mappings:
+ # There are some gaps, write a NULL entry so can lookup by index
+ line = ' NULL'
+ else:
+ line = ' "' + self.mappings[n] + '"'
+ # Add coma.
+ if n < self.highest_num:
+ line += ','
+ # Add number as aligned comment.
+ line += ' '*(90-len(line)) + '// ' + str(n)
+
+ self.f.write(line+'\n')
+
+ # End of array
+ self.f.write(' }\n')
+ # End of struct
+ self.f.write('};\n')
+ print('Re-generated', self.filename)
+
+ # Lookup function
+ self.f.write(LOOKUP_FUNCTION)
+
+ # Dump function
+ self.f.write(DUMP_FUNCTION)
+
+ # Add an individual mapping to the function
+ def addMapping(self, num, name):
+ # Handle some escapings
+ name = name.replace('\\', '\\\\')
+ name = name.replace('"', '""')
+
+ # Record.
+ self.mappings[num] = name
+ self.highest_num = num if num>self.highest_num else self.highest_num
+
+
+
+def main():
+ parser = argparse.ArgumentParser(description="Create the {} file.".format(ENTERPRISES_CFILE))
+ parser.add_argument('--infile')
+ parser.add_argument('outfile', nargs='?', default=ENTERPRISES_CFILE)
+ parsed_args = parser.parse_args()
+
+ # Read data from file or webpage
+ if parsed_args.infile:
+ with open(parsed_args.infile, encoding='utf-8') as fh:
+ data = fh.read()
+ else:
+ with urllib.request.urlopen(ENTERPRISE_NUMBERS_URL) as f:
+ if f.status != 200:
+ raise Exception("request for " + ENTERPRISE_NUMBERS_URL + " failed with result code " + f.status)
+ data = f.read().decode('utf-8')
+
+ # Find bits we need and generate enterprise entries
+ enterprises_content,last_updated_line = generate_enterprise_entries(data)
+
+ # Now write to a C file the contents (which is faster than parsing the global file at runtime).
+ c_file = CFile(parsed_args.outfile, last_updated_line)
+
+ mapping_re = re.compile(r'^(\d+)\s+(.*)$')
+ for line in enterprises_content.splitlines():
+ match = mapping_re.match(line)
+ if match:
+ num, name = match.group(1), match.group(2)
+ # Strip any comments and/or trailing whitespace
+ idx = name.find('#')
+ if idx != -1:
+ name = name[0:idx]
+ name = name.rstrip()
+ # Add
+ c_file.addMapping(int(num), name)
+
+
+
+if __name__ == "__main__":
+ main()