Find system-installed root SSL certificates on macOS (#14325)

FnControlOption · web-flow · commit e45b471ad358 · 2023-01-16T22:34:04.000Z
diff --git a/lib/std/crypto/Certificate/Bundle.zig b/lib/std/crypto/Certificate/Bundle.zig
@@ -60,13 +60,13 @@ pub fn rescan(cb: *Bundle, gpa: Allocator) !void {
         .windows => {
             // TODO
         },
-        .macos => {
-            // TODO
-        },
+        .macos => return rescanMac(cb, gpa),
         else => {},
     }
 }
 
+pub const rescanMac = @import("Bundle/macos.zig").rescanMac;
+
 pub fn rescanLinux(cb: *Bundle, gpa: Allocator) !void {
     // Possible certificate files; stop after finding one.
     const cert_file_paths = [_][]const u8{
@@ -195,25 +195,29 @@ pub fn addCertsFromFile(cb: *Bundle, gpa: Allocator, file: fs.File) !void {
         const decoded_start = @intCast(u32, cb.bytes.items.len);
         const dest_buf = cb.bytes.allocatedSlice()[decoded_start..];
         cb.bytes.items.len += try base64.decode(dest_buf, encoded_cert);
-        // Even though we could only partially parse the certificate to find
-        // the subject name, we pre-parse all of them to make sure and only
-        // include in the bundle ones that we know will parse. This way we can
-        // use `catch unreachable` later.
-        const parsed_cert = try Certificate.parse(.{
-            .buffer = cb.bytes.items,
-            .index = decoded_start,
-        });
-        if (now_sec > parsed_cert.validity.not_after) {
-            // Ignore expired cert.
-            cb.bytes.items.len = decoded_start;
-            continue;
-        }
-        const gop = try cb.map.getOrPutContext(gpa, parsed_cert.subject_slice, .{ .cb = cb });
-        if (gop.found_existing) {
-            cb.bytes.items.len = decoded_start;
-        } else {
-            gop.value_ptr.* = decoded_start;
-        }
+        try cb.parseCert(gpa, decoded_start, now_sec);
+    }
+}
+
+pub fn parseCert(cb: *Bundle, gpa: Allocator, decoded_start: u32, now_sec: i64) !void {
+    // Even though we could only partially parse the certificate to find
+    // the subject name, we pre-parse all of them to make sure and only
+    // include in the bundle ones that we know will parse. This way we can
+    // use `catch unreachable` later.
+    const parsed_cert = try Certificate.parse(.{
+        .buffer = cb.bytes.items,
+        .index = decoded_start,
+    });
+    if (now_sec > parsed_cert.validity.not_after) {
+        // Ignore expired cert.
+        cb.bytes.items.len = decoded_start;
+        return;
+    }
+    const gop = try cb.map.getOrPutContext(gpa, parsed_cert.subject_slice, .{ .cb = cb });
+    if (gop.found_existing) {
+        cb.bytes.items.len = decoded_start;
+    } else {
+        gop.value_ptr.* = decoded_start;
     }
 }
 
diff --git a/lib/std/crypto/Certificate/Bundle/macos.zig b/lib/std/crypto/Certificate/Bundle/macos.zig
@@ -0,0 +1,136 @@
+const std = @import("std");
+const assert = std.debug.assert;
+const mem = std.mem;
+const fs = std.fs;
+const Allocator = std.mem.Allocator;
+const Bundle = @import("../Bundle.zig");
+
+pub fn rescanMac(cb: *Bundle, gpa: Allocator) !void {
+    const file = try fs.openFileAbsolute("/System/Library/Keychains/SystemRootCertificates.keychain", .{});
+    defer file.close();
+
+    const bytes = try file.readToEndAlloc(gpa, std.math.maxInt(u32));
+    defer gpa.free(bytes);
+
+    var stream = std.io.fixedBufferStream(bytes);
+    const reader = stream.reader();
+
+    const db_header = try reader.readStructBig(ApplDbHeader);
+    assert(mem.eql(u8, "kych", &@bitCast([4]u8, db_header.signature)));
+
+    try stream.seekTo(db_header.schema_offset);
+
+    const db_schema = try reader.readStructBig(ApplDbSchema);
+
+    var table_list = try gpa.alloc(u32, db_schema.table_count);
+    defer gpa.free(table_list);
+
+    var table_idx: u32 = 0;
+    while (table_idx < table_list.len) : (table_idx += 1) {
+        table_list[table_idx] = try reader.readIntBig(u32);
+    }
+
+    const now_sec = std.time.timestamp();
+
+    for (table_list) |table_offset| {
+        try stream.seekTo(db_header.schema_offset + table_offset);
+
+        const table_header = try reader.readStructBig(TableHeader);
+
+        if (@intToEnum(TableId, table_header.table_id) != TableId.CSSM_DL_DB_RECORD_X509_CERTIFICATE) {
+            continue;
+        }
+
+        var record_list = try gpa.alloc(u32, table_header.record_count);
+        defer gpa.free(record_list);
+
+        var record_idx: u32 = 0;
+        while (record_idx < record_list.len) : (record_idx += 1) {
+            record_list[record_idx] = try reader.readIntBig(u32);
+        }
+
+        for (record_list) |record_offset| {
+            try stream.seekTo(db_header.schema_offset + table_offset + record_offset);
+
+            const cert_header = try reader.readStructBig(X509CertHeader);
+
+            try cb.bytes.ensureUnusedCapacity(gpa, cert_header.cert_size);
+
+            const cert_start = @intCast(u32, cb.bytes.items.len);
+            const dest_buf = cb.bytes.allocatedSlice()[cert_start..];
+            cb.bytes.items.len += try reader.readAtLeast(dest_buf, cert_header.cert_size);
+
+            try cb.parseCert(gpa, cert_start, now_sec);
+        }
+    }
+}
+
+const ApplDbHeader = extern struct {
+    signature: @Vector(4, u8),
+    version: u32,
+    header_size: u32,
+    schema_offset: u32,
+    auth_offset: u32,
+};
+
+const ApplDbSchema = extern struct {
+    schema_size: u32,
+    table_count: u32,
+};
+
+const TableHeader = extern struct {
+    table_size: u32,
+    table_id: u32,
+    record_count: u32,
+    records: u32,
+    indexes_offset: u32,
+    free_list_head: u32,
+    record_numbers_count: u32,
+};
+
+const TableId = enum(u32) {
+    CSSM_DL_DB_SCHEMA_INFO = 0x00000000,
+    CSSM_DL_DB_SCHEMA_INDEXES = 0x00000001,
+    CSSM_DL_DB_SCHEMA_ATTRIBUTES = 0x00000002,
+    CSSM_DL_DB_SCHEMA_PARSING_MODULE = 0x00000003,
+
+    CSSM_DL_DB_RECORD_ANY = 0x0000000a,
+    CSSM_DL_DB_RECORD_CERT = 0x0000000b,
+    CSSM_DL_DB_RECORD_CRL = 0x0000000c,
+    CSSM_DL_DB_RECORD_POLICY = 0x0000000d,
+    CSSM_DL_DB_RECORD_GENERIC = 0x0000000e,
+    CSSM_DL_DB_RECORD_PUBLIC_KEY = 0x0000000f,
+    CSSM_DL_DB_RECORD_PRIVATE_KEY = 0x00000010,
+    CSSM_DL_DB_RECORD_SYMMETRIC_KEY = 0x00000011,
+    CSSM_DL_DB_RECORD_ALL_KEYS = 0x00000012,
+
+    CSSM_DL_DB_RECORD_GENERIC_PASSWORD = 0x80000000,
+    CSSM_DL_DB_RECORD_INTERNET_PASSWORD = 0x80000001,
+    CSSM_DL_DB_RECORD_APPLESHARE_PASSWORD = 0x80000002,
+    CSSM_DL_DB_RECORD_USER_TRUST = 0x80000003,
+    CSSM_DL_DB_RECORD_X509_CRL = 0x80000004,
+    CSSM_DL_DB_RECORD_UNLOCK_REFERRAL = 0x80000005,
+    CSSM_DL_DB_RECORD_EXTENDED_ATTRIBUTE = 0x80000006,
+    CSSM_DL_DB_RECORD_X509_CERTIFICATE = 0x80001000,
+    CSSM_DL_DB_RECORD_METADATA = 0x80008000,
+
+    _,
+};
+
+const X509CertHeader = extern struct {
+    record_size: u32,
+    record_number: u32,
+    unknown1: u32,
+    unknown2: u32,
+    cert_size: u32,
+    unknown3: u32,
+    cert_type: u32,
+    cert_encoding: u32,
+    print_name: u32,
+    alias: u32,
+    subject: u32,
+    issuer: u32,
+    serial_number: u32,
+    subject_key_identifier: u32,
+    public_key_hash: u32,
+};
diff --git a/lib/std/io/reader.zig b/lib/std/io/reader.zig
@@ -3,6 +3,7 @@ const math = std.math;
 const assert = std.debug.assert;
 const mem = std.mem;
 const testing = std.testing;
+const native_endian = @import("builtin").target.cpu.arch.endian();
 
 pub fn Reader(
     comptime Context: type,
@@ -351,6 +352,14 @@ pub fn Reader(
             return res[0];
         }
 
+        pub fn readStructBig(self: Self, comptime T: type) !T {
+            var res = try self.readStruct(T);
+            if (native_endian != std.builtin.Endian.Big) {
+                mem.byteSwapAllFields(T, &res);
+            }
+            return res;
+        }
+
         /// Reads an integer with the same size as the given enum's tag type. If the integer matches
         /// an enum tag, casts the integer to the enum tag and returns it. Otherwise, returns an error.
         /// TODO optimization taking advantage of most fields being in order
diff --git a/src/link/MachO/fat.zig b/src/link/MachO/fat.zig
@@ -1,9 +1,7 @@
 const std = @import("std");
-const builtin = @import("builtin");
 const log = std.log.scoped(.archive);
 const macho = std.macho;
 const mem = std.mem;
-const native_endian = builtin.target.cpu.arch.endian();
 
 pub fn decodeArch(cputype: macho.cpu_type_t, comptime logError: bool) !std.Target.Cpu.Arch {
     const cpu_arch: std.Target.Cpu.Arch = switch (cputype) {
@@ -19,23 +17,13 @@ pub fn decodeArch(cputype: macho.cpu_type_t, comptime logError: bool) !std.Targe
     return cpu_arch;
 }
 
-fn readFatStruct(reader: anytype, comptime T: type) !T {
-    // Fat structures (fat_header & fat_arch) are always written and read to/from
-    // disk in big endian order.
-    var res = try reader.readStruct(T);
-    if (native_endian != std.builtin.Endian.Big) {
-        mem.byteSwapAllFields(T, &res);
-    }
-    return res;
-}
-
 pub fn getLibraryOffset(reader: anytype, cpu_arch: std.Target.Cpu.Arch) !u64 {
-    const fat_header = try readFatStruct(reader, macho.fat_header);
+    const fat_header = try reader.readStructBig(macho.fat_header);
     if (fat_header.magic != macho.FAT_MAGIC) return 0;
 
     var fat_arch_index: u32 = 0;
     while (fat_arch_index < fat_header.nfat_arch) : (fat_arch_index += 1) {
-        const fat_arch = try readFatStruct(reader, macho.fat_arch);
+        const fat_arch = try reader.readStructBig(macho.fat_arch);
         // If we come across an architecture that we do not know how to handle, that's
         // fine because we can keep looking for one that might match.
         const lib_arch = decodeArch(fat_arch.cputype, false) catch |err| switch (err) {
